ISSN 1854-6250 APEM journal Advances in Production Engineering & Management Volume 14 | Number 3 | September 2019 Published by CPE apem-journal.org University of Maribor Advances in Production Engineering & Management Identification Statement APEM ISSN 1854-6250 | Abbreviated key title: Adv produc engineer manag | Start year: 2006 ISSN 1855-6531 (on-line) Published quarterly by Chair of Production Engineering (CPE), University of Maribor Smetanova ulica 17, SI - 2000 Maribor, Slovenia, European Union (EU) Phone: 00386 2 2207522,Fax: 00386 2 2207990 Language of text: English APEM homepage: apem-journal.org UniversityofMaribor University homePage: WWW.um.si APEM Editorial Editor-in-Chief Miran Brezocnik editor@apem-journal.org, info@apem-journal.org University of Maribor, Faculty of Mechanical Engineering Smetanova ulica 17, SI - 2000 Maribor, Slovenia, EU Desk Editor Martina Meh desk1@apem-journal.org Janez Gotlih desk2@apem-journal.org Website Technical Editor Lucija Brezocnik desk3@apem-journal.org Editorial Board Members Eberhard Abele, Technical University of Darmstadt, Germany Bojan Acko, University of Maribor, Slovenia Joze Balic, University of Maribor, Slovenia Agostino Bruzzone, University of Genoa, Italy Borut Buchmeister, University of Maribor, Slovenia Ludwig Cardon, Ghent University, Belgium Nirupam Chakraborti, Indian Institute of Technology, Kharagpur, India Edward Chlebus, Wroclaw University of Technology, Poland Franci Cus, University of Maribor, Slovenia Igor Drstvensek, University of Maribor, Slovenia Illes Dudas, University of Miskolc, Hungary Mirko Ficko, University of Maribor, Slovenia Vlatka Hlupic, University of Westminster, UK David Hui, University of New Orleans, USA Pramod K. Jain, Indian Institute of Technology Roorkee, India Isak Karabegovic, University of Bihac, Bosnia and Herzegovina Janez Kopac, University of Ljubljana, Slovenia Qingliang Meng, Jiangsu University of Science and Technology, China Lanndon A. Ocampo, Cebu Technological University, Philippines Iztok Palcic, University of Maribor, Slovenia Krsto Pandza, University of Leeds, UK Andrej Polajnar, University of Maribor, Slovenia Antonio Pouzada, University of Minho, Portugal Rajiv Kumar Sharma, National Institute of Technology, India Katica Simunovic, J. J. Strossmayer University of Osijek, Croatia Daizhong Su, Nottingham Trent University, UK Soemon Takakuwa, Nagoya University, Japan Nikos Tsourveloudis, Technical University of Crete, Greece Tomo Udiljak, University of Zagreb, Croatia Ivica Veza, University of Split, Croatia Limited Permission to Photocopy: Permission is granted to photocopy portions of this publication for personal use and for the use of clients and students as allowed by national copyright laws. This permission does not extend to other types of reproduction nor to copying for incorporation into commercial advertising or any other profit-making purpose. Subscription Rate: 120 EUR for 4 issues (worldwide postage included); 30 EUR for single copies (plus 10 EUR for postage); for details about payment please contact: info@apem-journal.org Cover and interior design: Miran Brezocnik Printed: Tiskarna Kostomaj, Celje, Slovenia Subsidizer: The journal is subsidized by Slovenian Research Agency Statements and opinions expressed in the articles and communications are those of the individual contributors and not necessarily those of the editors or the publisher. No responsibility is accepted for the accuracy of information contained in the text, illustrations or advertisements. Chair of Production Engineering assumes no responsibility or liability for any damage or injury to persons or property arising from the use of any materials, instructions, methods or ideas contained herein. Copyright © 2019 CPE, University of Maribor. All rights reserved. Advances in Production Engineering & Management is indexed and abstracted in the WEB OF SCIENCE (maintained by Clarivate Analytics): Science Citation Index Expanded, Journal Citation Reports - Science Edition, Current Contents - Engineering, Computing and Technology • Scopus (maintained by Elsevier) • Inspec • EBSCO: Academic Search Alumni Edition, Academic Search Complete, Academic Search Elite, Academic Search Premier, Engineering Source, Sales & Marketing Source, TOC Premier • ProQuest: CSA Engineering Research Database -Cambridge Scientific Abstracts, Materials Business File, Materials Research Database, Mechanical & Transportation Engineering Abstracts, ProQuest SciTech Collection • TEMA (DOMA) • The journal is listed in Ulrich's Periodicals Directory and Cabell's Directory journal University of Maribor Chair of Production Engineering (CPE) Advances in Production Engineering & Management Volume 14 | Number 3 | September 2019 | pp 267-402 Contents Scope and topics Dynamic scheduling in the engineer-to-order (ETO) assembly process by the combined immune algorithm and simulated annealing method Jiang, C.; Xi, J.T. A blockchain-based smart contract trading mechanism for energy power supply and 270 271 284 demand network Hu, W.; Hu, Y.W.; Yao, W.H.; Lu, W.Q.; Li, H.H.; Lv, Z.W. A novel multiple criteria decision-making approach based on fuzzy DEMATEL, fuzzy ANP and 297 fuzzy AHP for mapping collection and distribution centers in reverse logistics Ocampo, L.A.; Himang, C.M.; Kumar, A.; Brezocnik, M. Effect of aluminium and chromium powder mixed dielectric fluid on electrical discharge 323 machining effectiveness Modi, M.; Agarwal, G. Multi-objective scheduling of cloud manufacturing resources through the integration of 333 Cat swarm optimization and Firefly algorithm Du, Y.; Wang, J.L.; Lei, L. Evaluation of the sustainability of the micro-electrical discharge milling process 343 Pellegrini, G.; Ravasio, C. A novel approximate dynamic programming approach for constrained equipment 355 replacement problems: A case study Sadeghpour, H.; Tavakoli, A.; Kazemi, M.; Pooya, A. An integrated system for scheduling of processing and assembly operations with 367 fuzzy operation time and fuzzy delivery time Yang, M.S.; Ba, L.; Zheng, H.Y.; Liu, Y.; Wang, X.F.; He, J.Z.; Li, Y. Optimal timing of price change with strategic customers under demand uncertainty: 379 A real option approach Lee, Y.; Lee, J.P.; Kim, S. Influence of high dynamic range images on the accuracy of the photogrammetric 391 3D digitization: A case study Santosi, Z.; Budak, I.; Sokac, M.; Hadzistevic, M.; Vukelic, D. Calendar of events 400 Notes for contributors 401 Journal homepage: apem-journal.org ISSN 1854-6250 (print) ISSN 1855-6531 (on-line) ©2019 CPE, University of Maribor. All rights reserved. Scope and topics Advances in Production Engineering & Management (APEM journal) is an interdisciplinary refer-eed international academic journal published quarterly by the Chair of Production Engineering at the University of Maribor. The main goal of the APEM journal is to present original, high quality, theoretical and application-oriented research developments in all areas of production engineering and production management to a broad audience of academics and practitioners. In order to bridge the gap between theory and practice, applications based on advanced theory and case studies are particularly welcome. For theoretical papers, their originality and research contributions are the main factors in the evaluation process. General approaches, formalisms, algorithms or techniques should be illustrated with significant applications that demonstrate their applicability to real-world problems. Although the APEM journal main goal is to publish original research papers, review articles and professional papers are occasionally published. Fields of interest include, but are not limited to: Additive Manufacturing Processes Advanced Production Technologies Artificial Intelligence in Production Assembly Systems Automation Big Data in Production Computer-Integrate d M anufacturing Cutting and Forming Processes Decision Support Systems Deep Learning in Manufacturing Discrete Systems and Methodology e-Manufacturing Evolutionary Computation in Production Fuzzy Systems Human Factor Engineering, Ergonomics Industrial Engineering Industrial Processes Industrial Robotics Intelligent Manufacturing Systems Joining Processes Knowledge Management Logistics in Production Machine Learning in Production Machine Tools Machining Systems Manufacturing Systems Materials Science, Multidisciplinary Mechanical Engineering Mechatronics Metrology in Production Modelling and Simulation Numerical Techniques Operations Research Operations Planning, Scheduling and Control Optimisation Techniques Project Management Quality Management Risk and Uncertainty Self-Organizing Systems Statistical Methods Supply Chain Management Virtual Reality in Production 270 Advances in Production Engineering & Management Volume 14 | Number 3 | September 2019 | pp 271-283 https://doi.Org/10.14743/apem2019.3.327 ISSN 1854-6250 Journal home: apem-journal.org Original scientific paper Dynamic scheduling in the engineer-to-order (ETO) assembly process by the combined immune algorithm and simulated annealing method Jiang, C.a*, Xi, J.T.a aSchool of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai, P.R. China A B S T R A C T A R T I C L E I N F O With the increasing demand for customization, the engineer-to-order (ETO) production strategy plays an increasingly important role in today's manufacturing industry. The dynamic scheduling problem in ETO assembly process was investigated. We developed the mathematical model to represent the problem. In order to reduce rescheduling frequency, we introduced the concept of starting time deviation and improved the rolling horizon driven strategy. We proposed the hybrid algorithm combining immune algorithm (IA) and simulated annealing (SA) with the minimization of the rescheduling cost as the objective. The IA was designed as the global search process and the SA was introduced to improve the local searching ability. The scenario-based approach was used to model the disruptions affecting the tasks to be executed. Performance of the rolling horizon driven strategy and the hybrid algorithm were evaluated through simulations, the experiment analysis showed the best parameters of rolling horizon methods and demonstrated the feasibility of the hybrid algorithm. The hybrid algorithm was tested on different scale benchmark instances and the case that collected from a steam turbine assembly shop. The quality of solution in terms of cost obtained by the hybrid algorithm was found superior to the other three algorithms proposed in the literature. © 2019 CPE, University of Maribor. All rights reserved. Keywords: Engineer-to-order (ETO); Assembly process; Dynamic scheduling; Rescheduling; Rolling horizon; Immune algorithm; Simulated annealing *Corresponding author: sjjzxjc@163.com (Jiang, C.) Article history: Received 16 April 2019 Revised 14 September 2019 Accepted 16 September 2019 1. Introduction The today's manufacturing industry is characterized by increasingly complex customized product, many companies have changed their business models from make-to-stock to X-to-order, where X usually stands for configure or engineer. Configure-to-order (CTO) is a modular approach, assembles orders from existing building blocks that can be delivered from stock and is hardly any engineering involved. The engineer-to-order (ETO) strategy is used when complex structures are needed to be built. The finished product and many components have never been built before and are impossible to be handled with standard variations. The ETO production strategy has become a trend [1,2]. During the ETO process, the most important phase is assembly process [3]. Assembly process accounts for almost 50 % of the total production time, 20 % of the total production cost and 30 % to 50 % of the labour cost [4]. The customer of ETO products are very strict with delivery deadline and late delivery will be punished. In ETO companies, several products are assembled in parallel [5] and the layout of typical ETO product assembly shop is shown in Fig. 1. Some workers formed the assembly group to execute the assembly task and the task duration depends on the number and skill level of workers [6]. The typical working situation of ETO assembly is shown in Fig. 2. 271 Jiang, Xi Fig. 1 The layout of typical ETO product assembly shop Fig. 2 Typical working situation of ETO assembly There are many unexpected events occur frequently during the ETO assembly process, such as the late delivery of necessary components, rework due to quality problems, etc [7,8]. There events usually exceed the task duration and cause the deviations from the initial schedule. Due to the characteristic of manual assembly and inherent flexibilities in the schedule, the small deviations can be absorbed by the initial schedule (The concept of 'small' depends on the question). If the deviation exceeds the threshold, the initial schedule became infeasible and rescheduling is needed. Since such unexpected events usually occur in the actual production, it is of practical significance to study the rescheduling problem in ETO assembly process. The terms 'rescheduling' and 'dynamic scheduling' are often used interchangeably in the literature [9]. Vieira et al. [10] have classified the dynamic scheduling strategies into three types: (1) reactive scheduling which is to repair the schedule [11]; (2) proactive scheduling which is to create a schedule robust with unexpected events [12]; (3) study how rescheduling strategy affect the performance of manufacturing systems. The dynamic scheduling problem in ETO assembly process is treated as reactive scheduling in multi-mode resource-constrained multi-project scheduling problem (MRCMPSP). For the review of reactive scheduling in project scheduling (PSP), we refer to Herroelen and Leus [12,13]. In the field of reactive scheduling in singlemode resource-constrained single-project scheduling problem(RCPSP), Van de Vonder [14] and Van de Vonder [15] et al. have dealt with the problem of task duration variability. The literature on reactive scheduling in the multi-mode resource-constrained single-project scheduling prob-lem(MRCPSP) is scare, Zhu proposed a branch-and-cut and constrained programming procedure for a general class of reactive problems [16], Deblaere proposed and evaluated a number of dedicated exact methods and tabu search to solve the reactive scheduling problem [11], Chakrabort-ty formulated the discrete time based models and proposed the reactive scheduling for a single or a set of disruptions [17]. To the best of our knowledge, the literature on reactive scheduling in the multi-mode resource-constrained multi-project scheduling problem (MRCMPSP) is none. The rescheduling problem is different from the initial planning problem because the rescheduling decision needed to be made in a timely manner. As the exact methods can only be used to solve small projects which have less than 20 tasks [18], the metaheuristics are more suitable [7,19]. Considering the convergence ability of the IA and the exploitation ability of SA, we proposed a hybrid algorithm which combined IA and SA to solve this problem. The rolling horizon rescheduling strategy is proposed to improve the computational speed [20]. The rest of this paper will be organized as the follows: Section 2 provides the mathematical model for the problem and section 3 describes the rolling horizon rescheduling strategy and the hybrid algorithm. In section 4, the hybrid algorithm and three other metaheuristic algorithms proposed in the literature are applied to solve benchmarks selected from literature and the industrial case. Conclusions and future directions are given in section 5. 2. Problem definition The decision problem concerns with rescheduling tasks and their corresponding worker allocation in the multi-project environment. The objective is to minimize the rescheduling cost, which is the sum of the task starting time deviation cost, mode switching cost and tardiness cost. The 272 Advances in Production Engineering & Management 14(3) 2019 Dynamic scheduling in the engineer-to-order (ETO) assembly process by the combined immune algorithm and simulated ... deviation of task starting time will incur the cost of additional storage and crane for the required components. The mode switching cost is often regarded as ''administrative'' cost [11]. The tardiness cost is the penalties associated with late project completion [21]. In this study, we consider the problem subject to the following assumptions: (1) Manual assembly processes are assumed to be carried out by a worker team. A mode represents a task-worker team with a constant duration. (2) During the execution of each task, the assigned mode cannot be changed, i.e. preemption is not allowed during the execution of each task. (3) The precedence relationships of each project force each task to be scheduled after all precedence tasks, the projects are independent of each other. (4) Each worker cannot be allocated to more than one task at the same time. (5) The maximum number of workers executing task is constrained by the work-space. (6) The set-up time for each task is included in the task duration, and the transportation time of workers between the tasks is negligible. (7) The rescheduled task starting time cannot be earlier than the task starting times from the initial schedule. The notation used in this section can be summarized as follows: Indices: Set of projects Set of tasks for project i e / Set of maximum number of tasks for each project that can be executed concurrently due to floor space constraint Set of task execution modes in task j e ]t, which correspond to the worker team Set of hierarchical levels Set of time periods Set of workers release date of project i e /, i.e. the earliest time that project i e / can start due date of project i e / the maximum number of tasks in project i that can be executed concurrently due to floor space constraint the predecessor set of task j e /¿, i.e. pred(j) = {j'\j' < j} maximum number of workers executing task j e ]t minimum number of workers executing task j e ]t number of type-fc workers number of type-fc workers in mode m e M^ duration of task j e ]t in mode m e M^ unit cost of starting time deviation of task j e ]t cost incurred by switching the mode from m^ to m* unit cost of violating the due date of project i start time of task j e ]t from the initial schedule processing time of rescheduled task j e ]t end time of rescheduled task j e ]t Decision variables: s*j start time of task j e ]t after rescheduling * _(1, if rescheduled task j is performed at time t e (0, T] lJt (0, otherwise * f 1, if rescheduled task j is executed in mode m e M^ Vijm "to,otherwise I Ji Q Mij K T W Parameters n di Ri pred(j) Wijmax Wijmin wk Z-mk dijm Kij Cíjm¡j Sij dur¡ f¡. Jii Advances in Production Engineering & Management 14(3) 2019 273 Jiang, Xi_ Under the assumptions and notations, the mathematical model is defined as follows: Objective function: ¡J ¡J i min C = XX Uij '< + ZZCijmi + Z Pi 'max(0, fi i=l7=1 ¿=17=1 ¿=1 Constraint conditions: s*j>ri ViEljEJi (2) K ytjm ' Zmk ^ Wijmax Vi e ',7 e Ji,m E Mij (3) ^íjmín k=1 / / ZZ Z'y¡Jm 'Zmk -Wk vt e (o'ne K (4) ¿ = 1 7 = 1 m=l Mij V + Z 'diJ'm Vi e 7,7 e e VredH) (5) m=l J ^xíi;.t0); Ep is the revenue of the backup unit. The model is subjected to the following constraints: Power difference balance: ^ Qi,P - ^ Qi,c + Qs = ienn ienr 0 (2) where, nc is the set of consumption units; Qi c is the power demand of the i-th consumption unit; Qs is the power consumption of the backup unit. If Qs > 0, the for-sale power is smaller than the power demand, and the gap needs to be eliminated by the backup unit; if Qs <0, the for-sale power is greater than the power demand, and the excess power needs to be consumed by the backup unit. The upper and lower bounds of the for-sale power provided by production units: Qi, p.min — Qí,p — Qi,p.max,^' e Hp (3) where, QiiP,min and Q(,p.max are the upper and lower bounds of the for-sale power provided by the Z-th production unit, respectively. Line transmission capacity: ^ Qi,pWxy,i + QsWxyii < Qxy, Vxy ei = 0 ienr, (4) where, Wxyi is the energy transfer distribution factor of node i to line xy; L is the set of all lines in the system; Qxy is the transmission capacity limit of line xy. The value of Wxy i can be determined by: Advances in Production Engineering & Management 14(3) 2019 287 Hu, Hu, Yao, Lu, Li, Lv located at the starting point 0, located at the branch (5) -1, located at the end point Eq. 5 gives the values of the energy transfer distribution factor at different conditions. The value of the factor is one if node i is located at the starting point of line xy, zero if node i is located at the branch of the line, and -1 if node i is located at the end point of the line. 3.3 Operating mechanism Our operating mechanism introduces blockchain smart contract into the traditional trading method, and implements the ERQ rule to ensure the self-sufficiency of the SDUs in the EPSDN, as well as maintaining the supply-demand balance, reducing costs and increasing revenue [9]. There are many advantages of this novel operating mechanism. For instance, the SDUs in the system can basically satisfy their own power demand for working and living through self-generation and the help from other units, which ensures the real-time balance of the grid; the social cost is cut down because no special supply is needed from the grid; the SDUs can acquire revenues from mutual assistance and transactions; the operation is efficient, secure and transparent, for the central node is replaced with point-to-point transactions between the SDUs; the supply and demand information is acquired timely, making it possible to consume the excess DG energies in the SDUs and eliminates the power difference in real time; the operating mechanism also promotes the development and use of clean energy, because most of the DG energies are clean in nature. The operating mechanism of the power trading system is illustrated in Fig. 4. Fig. 4 Operating mechanism of the power trading system 288 Advances in Production Engineering & Management 14(3) 2019 A blockchain-based smart contract trading mechanism for energy power supply and demand network The specific steps of the operating mechanism are as follows: Step 1: All the SDUs in the distribution network determine the day-ahead production and consumption plans, and carry out preliminary transactions in the primary power market. Step 2: An SDU will become a consumption unit if it demands more power than the planned production for the day, and will request other SDUs in the system to satisfy its demand. Step 3: Each nearby production unit accepting the request will calculate the total power demand and judge whether it can fully satisfy the demand. If yes, the production unit will directly submit the available power and quotation; otherwise, the production unit will request the backup unit to compensate for the gap and submit the quotation. Step 4: All quotations are cleared according to the ERQ rule: All valid quotations are ranked in ascending order until the power difference is eliminated. Step 5: The power transfer distribution factor is calculated to judge whether the power exceeds the transmission capacity limit of any line. If not, the security review is passed and the next step will be executed; if yes, the clearing queue and price will be modified to reduce the power transmitting through the line. The modification steps include (1) computing the capacity AQxy of the target line xy; (2) increasing the quotation of the production unit at the start point of the line until the line capacity is reduced to AQxy; (3) repeating Step 4 to confirm the transaction price and repeating Step 5 until the power does not exceed the transmission capacity limit of any line. Step 6: The production unit and backup unit participating in the transaction will arrange the output and load to ensure the supply-demand balance of the distribution network. Step 7: Both parties will settle according to the agreed quotation. 4. Efficient trading mechanism based on smart contract 4.1 Smart contract Blockchain, a revolutionary technology in the Internet era, adopts an underlying decentralized collaboration mechanism. The data are linked up by chronologically generated blocks, forming a data structure suitable for any decentralized trust network. The unique formation mechanism has blessed the blockchain with such features as decentralized, trust-free, traceable and smart contracted [17, 18]. In essence, a smart contract running on a blockchain is a computer program that is automatically executed according to certain rules [19, 20]. Once being reached between the transaction parties in the smart contract, the agreement will be automatically executed by the pre-written code and cannot be intervened. In this way, the contact can be signed and executed more efficiently at a lower. Similar to that of traditional contracts, the lifecycle of smart contract (Fig. 5) involves such three phases as contract generation, contract release and contract execution. Contract generation Contract release Contract execution Fig. 5 The lifecycle of smart contract Advances in Production Engineering & Management 14(3) 2019 289 Hu, Hu, Yao, Lu, Li, Lv In the phase of contract generation, the multiple parties need to negotiate over the contract, prepare and routinize contract text, and judge whether the program is consistent with the text. Firstly, the contracting parties should determine their rights and obligations through discussion, and initialize the draft of the contract. Secondly, relevant persons with professional knowledge and legal literacy will review the conformance and legal effect of the draft, and finalize the text in paper. Thirdly, professional technicians will routinize the contract text into a program based on the blockchain, and conduct a trial run on the virtual machine to check the consistency between the program and the text. If consistent, the next phase will be initiated; otherwise, the above steps need to be go through again. The contract release is relatively simple. After passing the program test, the contract can be released to all nodes in the network. The contract execution is based on a preset trigger condition. Once the condition is satisfied, all terms in the contract will be executed automatically in an open and transparent manner, and all transaction information will be recorded; otherwise, the program will be terminated immediately. 4.2 Trading mechanism based on smart contract The trading mechanism based on smart contract should carry the following three features: All SDUs are free to enter and exit the power trading market; the quotation of each production unit is kept confidential before clearing; the relevant terms in the contract should be executed automatically [21]. Here, the trading mechanism is divided into six phases, namely, request issuance, request acceptance, quotation sealing, determination of clearing queue and transaction price (CQTP determination), security review, and transaction settlement. The six phases are corresponding to six performance functions: request issuance function, request acceptance function, quotation sealing function, CQTP determination function, security review function and transaction settlement function. Next, the power difference purchase of a consumption unit was taken as an example to illustrate the execution of smart contract. The transaction process is shown in Fig. 6. 290 Fig. 6 Smart contract trading mechanism Advances in Production Engineering & Management 14(3) 2019 A blockchain-based smart contract trading mechanism for energy power supply and demand network During the request issuance, any SDU in the distribution network can issue a transaction request as a consumption unit, after transferring a certain amount of virtual currency to the smart contract address. The virtual currency serves as a deposit against false request. In the request acceptance phase, each SDU with excess power in the distribution network automatically becomes a production unit, calculates the total power demand of the consumption unit, and judges if it can satisfy the demand. If necessary, the backup unit will be started to compensate for the gap. The quotation sealing, the core of the trading mechanism, directly bears on the transaction fairness. Since the smart contract requires the production unit quotations to be kept confidential before clearing, the quotation process was divided into the sealed quotation stage and the public quotation stage. In the sealed quotation stage, each production unit connects its real quotation with random strings and performs hash encryption, because the hash function is easy to check and cannot be solved reversibly. The encrypted hash value is used as the sealed quotation and submitted before the deadline. This approach keeps the quotation unique and confidential. To prevent malignant competition, a certain amount of virtual currency should be transferred to the smart contract address as the deposit. Definition 1: The hash function hash:D ^ R represents the mapping from D to R. Let d = \D\ and r = |R| be the size of the definition domain and the value domain, respectively. Then, d >> r, i.e. the value range of the definition domain is much larger than that of the value domain. Thus, the has function is a "many-to-one" mapping that maps information of random sizes into a uniform size hash value. According to Definition 1, the quotation sealing function can be determined as: H = hash(t, a) (6) where, H is the sealed quotation; hash{•) is a hash function; t is the real quotation; a is a random string. The security of the hash-based quotation sealing function is demonstrated as follows: (1) the hash(t, a) can be computed easily for any given t and a; (2) it is infeasible to find hash(t,a) = H for any given H, due to the unidirectional nature of the hash function; (3) it is infeasible to find t' and a' such that hash(t, a) = hash(t', a') for any given t and a, due to the weak collision resistance of the hash function; (4) it is infeasible to find any pair of (t, a) and (t', a') such that hash(t, a) = hash(t', a'), due to the strong collision resistance of the hash function. According to the principle of cryptography, the quotation sealing function enjoys excellent security. In the CQTP phase, each production unit needs to submit the real quotation and the random string before the deadline of the public quotation. Then, the smart contract will check whether the hash function value hash(t, a) equals the sealed quotation H submitted in the previous stage. If not, the quotation will be discarded; otherwise, the clearing queue and transaction price will be determined by the ERQ rule. If the new quotation is high than the highest quotation in the clearing queue and satisfies the capacity requirement, the next phase will be kicked off; otherwise, the clearing queue will be updated repeatedly until this phase is completed. The security review is essential to any transaction. This phase mainly observes whether there is a physical unrealizable situation. This paper judges whether the adjustment power submitted by the transaction parties will exceed the transmission capacity limit of the corresponding line. If yes, adjustment should be made as per the method in Subsection 3.3 until the problem is solved. Once confirmed by the smart contract, the transaction volume and price of the two parties will be executed automatically and cannot be modified. The transaction settlement is relatively simple. In this phase, the two parties execute the transaction in strict accordance with the price and volume determined in the previous phase, as well as the transaction settlement rules. After the transaction is completed, the deposits will be returned to the relevant parties. Advances in Production Engineering & Management 14(3) 2019 291 Hu, Hu, Yao, Lu, Li, Lv 4.3 Transaction settlement rules Within the specified transaction period, all SDUs in the EPSDN involved in the transaction need to adjust their production and consumption plans, i.e. increasing the production or reducing the consumption, for the purpose of transaction [22, 23]. The transaction settlement platform observes the real-time production and consumption of each participant via smart meters, and makes settlement according to the specific situation. Firstly, the deposits paid by all production units that fail to win the bid will be returned. Then, the settlement of the bid-winning production unit will be carried out in three cases. Case 1: Supply-demand balance (the production unit can adjust the production and consumption plan according to the transaction result.) This is the most desirable outcome. In this case, the settlement should be carried out at the price agreed between the production and consumption units, and the deposits should be returned to the relevant parties. Case 2: Oversupply (the adjustment amount of the production unit exceeds the amount required for the transaction.) In this case, the output of the backup unit should be reduced to maintain the balance between supply and demand. The two parties should settle at the agreed price and volume. Then, the production unit should receive a compensation for the backup unit: where, C is the compensation amount; a is the compensation factor; psb is the cost of starting backup unit; AQ is the difference between the actual adjustment amount of the production unit and the amount required for the transaction; pge is the unit cost of power production; pts is the unit loss of the power transmission. Note that the value of a can be changed according to the transaction conditions, and is generally below 1. Case 3: Short supply (the adjustment amount of the production unit fails to reach the amount required for the transaction.) In this case, the output of the backup unit should be increase to maintain the balance between supply and demand. The two parties should settle at the agreed price and volume. Then, the production unit should be imposed a penalty for the backup unit: where, D is the penalty amount; /3 is the penalty factor; AQ is the difference between the actual adjustment amount of the production unit and the amount required for the transaction. Note that the value of f> can be changed according to the transaction conditions, and is generally above 1. 5. Results and discussion: A case study The proposed transaction mechanism was verified with an EPSDN containing 6 SDUs, denoted as SDUs A-F. The EPSDN structure is given is Fig. 7, where the nodes 3, 6, 7, 10, 13 and 15 correspond to the said SDUs. It is assumed that the day-ahead production plan has been determined, regardless of the participation of the standby unit. The remaining transmission capacity of each line in the distribution network is shown in Table 1. In the simulation test, the SDUs in need of more power, i.e. consumption units, should initiate a request every 20 min according to their own demands. The request contains the power demand after 20 min. The production units should determine whether to accept the request within 1 min after the issuance. Then, quotation sealing, CQTP determination, security review, and settlement should be completed within 5 min, 10 min, 15 min and 20 min, respectively. After 20 min, the bid-winning production unit should supply the agreed amount of power to the consumption unit. C = a-psb-AQ Psb = Pge ~Pts (7) (8) D=ß-psb-AQ' (9) 292 Advances in Production Engineering & Management 14(3) 2019 A blockchain-based smart contract trading mechanism for energy power supply and demand network Fig. 7 The energy power supply and demand network (EPSDN) structure Table 1 The remaining transmission capacity of each line in the distribution network Branch node Capacity remaining/kWh Branch node Capacity remaining/kWh 1-2 1.3 8-9 4.2 1-7 1.3 9-10 2.7 1-12 1.3 10-11 3.5 2-3 3.4 12-13 2.6 3-4 2.8 13-14 3.3 4-5 4.3 14-15 2.4 5-6 2.6 15-16 4.5 7-8 3.7 16-17 1.9 Before the simulation test, each SDU was given 5 units of virtual currency. The SDUs A, B and C were regarded as consumption units that send transaction requests to the system, while the SDUs D, E and F were considered as production units with adjustment ability that respond to the system requests. Ten transactions were simulated to fully verify the efficiency of the blockchain-based smart contract trading mechanism. Tables 2-4 respectively list the sealed quotation, real quotation and random string of each bidding production unit, the remaining capacity of each line before and after the security review, and the results of smart contract transaction. Fig. 8 provides the scatter plot on the transaction results. Table 2 Sealed quotation, real quotation and random string Stage 1 2 3 4 5 6 7 8 9 10 Qwnl23 wid234 QDFcgh4 Dgknvhr DFced- Xcdjsgfb ckdjCK- Cdfjsjnvl hdk- dfdkshD Sealed offer njdeijsw mmj832f 59oh90d 47jfi8SD mkfj736 84DCkch JHldkDC ai47hdfj cASD23k F675djfh 3wde guQDc b45d vbn 4mkddc 987 76cjf 9fghj f34kdkfd dhc Real offer(Virtual currency/kWh) Random string bgh xdr dfg knb okm lkj gfd esz ygv fer 1.2 1.5 1.9 2.3 2.7 3 3.5 3.8 4.2 4.9 Table 3 The remaining capacity of each line before and after the security review Stage 1 2 3 4 5 6 7 8 9 lo" Line 1-2 1-7 2-3 8-9 9-10 3-4 1-12 7-8 1011 1516 1617 3-4 8-9 5-6 4-5 1011 1213 1314 1-7 7-8 Remaining branch capacity before 1.3 1.3 3.4 4.2 2.7 2.8 1.3 3.7 3.5 4.5 1.9 2.8 4.2 2.6 4.3 3.5 2.6 3.3 1.3 3.7 safety audit/kWh Remaining branch capacity after 2.0 2.5 3.4 4.2 3.0 3.5 2.3 3.7 3.5 4.5 3.1 3.7 4.2 2.9 5.0 3.5 3.8 3.3 3.6 4.0 safety audit/kWh Advances in Production Engineering & Management 14(3) 2019 293 Hu, Hu, Yao, Lu, Li, Lv Table 4 The results of smart contract transaction Stage Total demand Supply and demand unit transaction Production unit Volume Transaction revenue for consumer units/kWh price (Virtual currency/kWh) of transaction (kWh) (Virtual currency) 1 0.8 1.582 D E 0.2 0.6 0.3164 0.9492 2 0.9 1.734 D F 0.4 0.5 0.6936 0.8670 3 0.3 2.012 E F D 0.2 0.1 0.2 0.4024 0.2012 0.4924 4 1.2 2.462 E F 0.6 0.4 1.4772 0.9848 5 0.7 2.813 D F 0.3 0.4 0.8439 1.1252 6 1.0 3.272 D E 0.3 0.7 0.9816 2.2904 7 0.5 3.657 E F 0.4 0.1 1.4628 0.3657 8 0.1 3.923 E 0.1 0.3923 9 0.6 4.502 D F 0.3 0.3 1.3506 1.3506 10 1.1 5.216 E F 0.6 0.5 3.1296 2.6080 3.5000 f-, § 3.0000 ■a 2.5000 £ 2.0000 5 1.5000 * £ 1.0000 • • . I * * * « 0.5000 _ • g ; i H • H 0.0000 O 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Volume ( kw.h > Fig. 8 The scatter plot on the transaction results The results show that the three production units submitted the quotation sealing functions within the specified time, converting real quotes into random strings of the same length to ensure fair and secure transactions. Determined the clearing queue and price through the intelligent contract-based ERQ rule, once the parties have reached an agreement, they will immediately follow the agreement to prevent transaction friction. It can be seen from Table 3 that after the security review, the remaining capacity of each branch has undergone a certain change, which illustrates the necessity of the security review and prevents the situation from exceeding the capacity limit. According to the final results of the 10 tests, the SDU D and the SDU E provided 0.8 kWh power to the consumption units in test 1, and respectively earned 0.3164 and 0.9492 units of virtual currently; the SDU E and the SDU F provided 0.3 kWh power to the consumption units in test 3, and respectively earned 0.4024 and 0.2012 units of virtual currently; the SDU D and the SDU F provided 0.7 kWh power to the consumption units in test 5, and respectively earned 0.8439 and 1.1252 units of virtual currently. The SDU E and the SDU F provided 0.5 kWh power to the consumption units in test 7, and respectively earned 1.4628 and 0.3657 units of virtual currently. the SDU D and the SDU F provided 0.6 kWh power to the consumption units in test 9, and respectively earned 1.3506 and 1.3506 units of virtual currently. In general, the SDUs in the network completed the targets in a secure and efficient manner: the SDUs A, B and C fulfilled their 294 Advances in Production Engineering & Management 14(3) 2019 A blockchain-based smart contract trading mechanism for energy power supply and demand network power demands, while the SDUs D, E and F respectively received 4.6785, 10.1039 and 7.5025 units of virtual currency. The scatter plot on the transaction results clearly displays that the transaction revenue basically increased with the power amount of the transaction, except for some singularities (i.e. the points with low power amount and high revenue). The trend reveals the importance of the quotation sealing of the production units, and that correct quotation can lead to better revenue. 6. Conclusion In DG-dominated distribution networks, the traditional transaction method is prone to malicious attacks, which threatens the transaction security, and troubled by high transaction cost and poor transparency. To solve these problems, this paper sets up a point-to-point, secure and efficient trading system based on blockchain and smart contract. Firstly, the ERQ rule was adopted to maximize the revenue of the power trading market under the constraints on power difference, upper and lower bounds of for-sale power provided by production units, and the line transmission capacity, yielding a clearing queue and transaction price. On this basis, the author detailed the specific steps and implementation of the efficient power trading system. Considering the complexity of point-to-point transactions, the transaction was divided into six phases based on blockchain smart contract, namely, request issuance, request acceptance, quotation sealing, the CQTS determination, security review and settlement, aiming to ensure the safe and efficient operation of transactions in the distribution network. Through example analysis, the proposed method was proved capable of improving the transaction security and transparency in small-scale power markets involving numerous traders. Using the method proposed in this paper can help large and medium-sized enterprises to sell excess electricity or purchase the lack of electricity, maintain the balance of electricity consumption, thereby improving the economic efficiency of enterprises. There are many other areas worth exploring for the integration between blockchain and energy trading, such as security and economic analysis on blockchain-based energy trading (i.e. optimizing the security and cost efficiency of energy trading in a decentralized and trust-free blockchain environment) and the construction of smart contract trading system for the fully open power market. Hence, the future research will establish optimization models for the cost efficiency of power transactions under the blockchain-based energy trading system, trying to further promote the application of our trading mechanism. Acknowledgement This paper is made possible thanks to the generous support from the National Social Science Foundation of China (No.19BGL003). References [1] Liu, Z. (2015). Global energy internet, China Electric Power Press, Beijing, China. [2] Huckle, S., Bhattacharya, R., White, M., Beloff, N. (2016). Internet of things, blockchain and shared economy applications, Procedia Computer Science, Vol. 98, 461-466, doi: 10.1016/j.procs.2016.09.074. [3] Sikorski, J.J., Haughton, J., Kraft, M. (2017). Blockchain technology in the chemical industry: Machine-to-machine electricity market, Applied Energy, Vol. 195, 234-246, doi: 10.1016/j.apenergy.2017.03.039. [4] Chiu, Y.-S.P., Chen, H.-Y., Chiu, T., Chiu, S.W. (2018). Incorporating flexible fabrication rate and random scrap into a FPR-based supply-chain system, Economic Computation and Economic Cybernetics Studies and Research, Vol. 52, No. 2, 157-174, doi: 10.24818/18423264/52.2.18.10. [5] Sharma, P.K., Moon, S.Y., Park, J.H. (2017). Block-VN: A distributed blockchain based vehicular network architecture in smart city, Journal of Information Processing Systems, Vol. 13, No. 1, 184-195, doi: 10.3745/IIPS.03.0065. [6] Lalami, I., Frein, Y., Gayon, J.P. (2017). Demand variability and value of information sharing in the supply chain. A case study in the automotive industry, Journal Européen des Systèmes Automatisés, Vol. 50, No. 1-2, 157-186, doi: 10.3166/JESA.50.157-186. [7] Kshetri, N. (2017). Can blockchain strengthen the internet of things?, IT Professional, Vol. 19, No. 4, 68-72, doi: 10.1109/MITP.2017.3051335. Advances in Production Engineering & Management 14(3) 2019 295 Hu, Hu, Yao, Lu, Li, Lv [8] Hussein, A.F., Arunkumar, N., Ramirez-Gonzalez, G., Abdulhay, E., Tavares, J.M.R.S., De Albuquerque, V.H.C. (2018). A medical records managing and securing blockchain based system supported by a genetic algorithm and discrete wavelet transform, Cognitive Systems Research, Vol. 52, 1-11, doi: 10.1016/j.cogsys.2018.05.004. [9] Deng, J., Wang, H.-B., Wang, C.-M., Zhang, G.-W. (2017). A novel power market clearing model based on the equilibrium principle in microeconomics, Journal of Cleaner Production, Vol. 142, Part 2, 1021-1027, doi: 10.1016/ j.jclepro.2016.08.146. [10] Li, B., Lu, C., Cao, W., Qi, B., Li, D., Chen, S., Cui, G. (2017). Application of automatic demand response system based on blockchain technology, Chinese Journal of Electrical Engineering, Vol. 37, No. 13, 3691-3702, doi: 10.13334/j.0258-8013.pcsee.162462. [11] Ai, X., Yang, M., Liu, Z., Li, X. (2017). Modelling and control safety of digital push-pull switched mode power supply, European Journal of Electrical Engineering, Vol. 19, No. 5-6, 341-355. [12] Huckle, S., White, M. (2017). Fake news: A technological approach to proving the origins of content, using block-chains, Big Data, Vol. 5, No. 4, 356-371, doi: 10.1089/big.2017.0071. [13] Hoy, M.B. (2017). An introduction to the blockchain and its implications for libraries and medicine, Medical Reference Services Quarterly, Vol. 36, No. 3, 273-279, doi: 10.1080/02763869.2017.1332261. [14] Benchoufi, M., Porcher, R., Ravaud, P. (2017). Blockchain protocols in clinical trials: Transparency and traceabil-ity of consent, F1000 Research, Vol. 6, No. 66, doi: 10.12688/f1000research.10531.3. [15] Suda, M., Tejblum, B., Francisco, A. (2017). Chain reactions: Legislative and regulatory initiatives related to blockchain in the United States, Computer Law Review International, Vol. 18, No. 4, 97-103, doi: 10.9785/cri-2017-0402. [16] Roth, U., Ngne Djoua, T. (2018). Message exchange on base of a blockchain-based layered architecture, IT-Information Technology, Vol. 60, No. 5-6, 253-261, doi: 10.1515/itit-2017-0037. [17] Fu, Y., Zhu, J., Gao, S. (2018). CPS information security risk evaluation based on blockchain and big data, Tehnicki Vjesnik - Technical Gazette, Vol. 25, No. 6, 1843-1850, doi: 10.17559/TV-20180621055030. [18] Lu, Y. (2019). The blockchain: State-of-the-art and research challenges, Journal of Industrial Information Integration, Vol. 15, 80-90, doi: 10.1016/j.jii.2019.04.002. [19] Jegou, D., Kim, D.-W., Baptiste, P., Lee, K.H. (2006). A contract net based intelligent agent system for solving the reactive hoist scheduling problem, Expert Systems with Applications, Vol. 30, No. 2, 156-167, doi: 10.1016/j.eswa. 2005.06.019. [20] He, H.W., Yan, A., Chen, Z.H. (2018). Survey of smart contract technology and application based on blockchain, Journal of Computer Research and Development, Vol. 55, No. 11, 2452-2466, doi: 10.7544/issn1000-1239.2018. 20170658. [21] Ping, J., Chen, S., Zhang, N., Yan, Z., Yao, L.Z. (2017). Decentralized transactive mechanism in distribution network based on smart contract, In: Proceedings of the Chinese Society of Electrical Engineering (CSEE), Vol. 37, No. 13, 3682-3690, doi: 10.13334/j.0258-8013.pcsee. 170374. [22] Althaher, S., Mancarella, P., Mutale, J. (2015). Automated demand response from home energy management system under dynamic pricing and power and comfort constraints, IEEE Transactions on Smart Grid, Vol. 6, No. 4, 1874-1883, doi: 10.1109/TSG.2014.2388357. [23] Yan, H., Li, B., Chen, S., Zhong, M., Li, D., Jiang, L., He, G. (2015). Future evolution of automated demand response system in smart grid for low-carbon economy, Journal of Modern Power Systems and Clean Energy, Vol. 3, No. 1, 72-81, doi: 10.1007/s40565-015-0103-5. Appendix A Used abbreviations: EPSDN Energy power supply and demand network ERQ Encourage-real-quotation SDU Supply and demand unit DG Distributed generation CQTP Clearing queue and transaction price 296 Advances in Production Engineering & Management 14(3) 2019 Advances in Production Engineering & Management Volume 14 | Number 3 | September 2019 | pp 297-322 https://doi.Org/10.14743/apem2019.3.329 ISSN 1854-6250 Journal home: apem-journal.org Original scientific paper A novel multiple criteria decision-making approach based on fuzzy DEMATEL, fuzzy ANP and fuzzy AHP for mapping collection and distribution centers in reverse logistics Ocampo, L.A.a*, Himang, C.M.b, Kumar, A.c, Brezocnik, M.d department of Industrial Engineering, Cebu Technological University, Cebu City, Philippines bGraduate School, Cebu Technological University, Cebu City, Philippines cCentre for Supply Chain Improvement, University of Derby, United Kingdom dFaculty of Mechanical Engineering, Intelligent Manufacturing Systems Laboratory, University of Maribor, Maribor, Slovenia A B S T R A C T A R T I C L E I N F O The strategic location of reverse logistics facilities enables organizations to obtain optimal performance to collect end-of-line (EOL) products and distribute remanufactured products effectively and efficiently. The planning of facility location entails consideration of multiple essential criteria rather than optimizing a single criterion. This paper develops a methodological framework based on an integrated multiple criteria decision-making (MCDM) approach that captures the complexity of location planning for collection and distribution centers under fuzzy conditions utilizing decision making trial and evaluation laboratory (DEMATEL), analytic network process (ANP), and analytic hierarchy process (AHP). This novel approach aids decision-makers to simultaneously select a separate location for collection and distribution through a holistic assessment of a location's viability for both purposes. It advances the reverse logistics literature by considering multiple criteria and their interrelationships in the location selection process, along with uncertainty and vagueness in decision making. Additionally, the proposed approach allows flexibility for decision-makers as they retain the control in picking a site based on its priority on being a collection or distribution center. Results show that government policies and regulations play a vital role in the facility location decision as they interact mostly with other criteria. Moreover, results also suggest that quantity and quality uncertainties for remanufacturing are significant factors that must be taken into consideration in the collection function, while economic and market-oriented issues are major concerns for a distribution function. This finding was observed through the application of the proposed methodological framework in a case study of the furniture industry in the Philippines. The practical implications of this study focus on being an aid in organizing and improving the operations of the reverse logistics sector of the Philippines. Finally, the proposed approach can be used to address general facility location problems in other industrial applications where tradeoffs among stakeholders or entities are well pronounced and decision-makers find it imperative that such tradeoffs must be carefully considered. © 2019 CPE, University of Maribor. All rights reserved. Keywords: Reverse logistics; Collection; Distribution; Fuzzy environment; Remanufacturing; Multiple criteria decision-making (MCDM); Decision-making and trial evaluation laboratory (DEMATEL); Analytic network process (ANP); Analytic hierarchy process (AHP) Corresponding author: lanndonocampo@gmail.com (Ocampo, L.A.) Article history: Received 13 November 2018 Revised 9 September 2019 Accepted 12 September 2019 1. Introduction Sustainable practices have become a continual pursuit of manufacturers to address prevalent issues on environmental awareness, resource depletion, consumer awareness of sustainability impacts, legislation, corporate imaging, economic benefits, and government incentives (Mutha and Pokharel, 2009 [1]; Sheu, 2011 [2]; Rashid et al, 2013 [3]; Govindan et al, 2015 [4, 5]). One 297 Ocampo, Himang, Kumar, Brezocnik chief sustainable practice is the end of life (EOL) strategies which intend to restore goods to its original working condition (USITC, 2012 [6]). Among the extant EOL strategies, the concept of remanufacturing is of growing interest to scholars from domain disciplines (Rashid et al., 2013 [3]). As an industrial process, the goal of remanufacturing is to recover the residual value of used products by reconditioning and reusing components that are still functional and acceptable (Wei et al., 2015 [7]). It is a product recovery technique (PRT) that promotes sustainability as it helps firms achieve closed-loop supply chains. Remanufacturing addresses the environmental, social, and economic dimensions of sustainability by minimizing waste and emission generation, creating jobs, and trimming down production costs by 50 % (Rathore et al., 2011 [8]; Chen and Chang 2012 [9]; Xiaoyan, 2012 [10]). Several original equipment manufacturers (OEMs) have taken an interest in remanufacturing such as Caterpillar, HP, Xerox, and Kodak to increase profit and improve their social and environmental performances as well. This increased attention can be attributed to remanufacturing's benefits and essential functions in the ever-changing society. In remanufacturing, one of the crucial aspects is reverse logistics. Reverse logistics is the process of planning, implementing, and controlling efficient, effective inbound flow, inspection, and disposition of returned products and related information for recovering value (Srivastava, 2006 [11]). The collected EOL products are subjected to a detailed inspection, which either ends up remanufactured or disposed. Products that go through the remanufacturing process are distributed in secondary markets; afterward, the cycle of collection and remanufacturing continues. The practice of remanufacturing, however, is rather hindered despite its advantages in terms of sustainability since the receptivity of consumers varies from one region to another, as suggested in the current literature. That is, consumers in well-developed Western countries are more open to remanufactured products compared to those in most developing countries (Nnorom and Osi-banjo, 2008 [12]; Zou et al., 2016 [13]). As critical tasks of reverse logistics, several studies in the literature tackled how these functions can be optimized according to collection rate and sales (Malik et al., 2015 [14]; Pop et al., 2015 [15]), profit and return rate (Hong and Yeh, 2012 [16]), and economies of scale (Atasu et al., 2013 [17]), to name a few. Consequently, dominant mathematical models such as continuous modeling frameworks (Wojanowski et al., 2007 [18]), a mixed-integer nonlinear model (Min and Ko, 2008 [19]), and graph theory and matrix approach (Malik et al., 2015 [14]) are adapted to design such functions. While prior studies in literature present mathematical models with single objective analyses to optimize collection and distribution decision problems, these methodologies fail to incorporate various aspects and holistic considerations that are necessary for the decision problem involving the location of centers (Malik et al., 2015 [14]). Real-world problems are rarely single objective but are multi-objective; therefore, multi-objective approaches should be given more attention and focus (Govindan et al., 2015 [4, 5]). Additionally, results are expected to be more informed, and better decisions are drawn when an appropriate structure of the problem and evaluation of the multi-criteria nature of the problem is explicitly established. Hence, multi-criteria decision-making (MCDM) approaches are introduced in the current literature. In the field of remanufacturing, pertinent issues are successfully resolved using MCDM approaches such as: identifying a strategic model for distribution channel management using fuzzy analytical hierarchy process (FAHP) and hierarchical fuzzy technique for order of preference and similarity to ideal solutions (HFTOPSIS) (Paksoy et al., 2012 [20]), analyzing the interrelationships between risks faced by third-party logistics service providers (3PLs) using decision-making and trial evaluation laboratory (DEMATEL) (Govindan and Chaudhuri, 2016 [21]), and selecting important criteria in considering factors of reverse logistics implementation using FAHP (Chiou et al., 2012 [22]), to name a few. Given that the selection of a logistics center can be modeled as a decision problem that involves critical elements and that an integrated approach of simultaneously selecting distribution and collection centers lacks in the current literature, this paper aims to simultaneously identify a location for collection and distribution centers using MCDM approach. With an MCDM model, complexity and uncertainty of the selection process may mimic real-life decision-making with different and contradictory criteria and alternatives. Further, it is imperative to recognize that while the selection of collection and distribution centers are addressed in separate conditions, 298 Advances in Production Engineering & Management 14(3) 2019 A novel multiple criteria decision-making approach based on fuzzy DEMATEL, fuzzy ANP and fuzzy AHP for mapping ... the need to simultaneously resolve both logistic centers remains relevant in the context of economic and operational sustainability. Thus, this paper attempts to map both collection and distribution centers simultaneously using an integrated MCDM approach consists of fuzzy DEMATEL, fuzzy analytic network process (FANP), and FAHP. The proposed approach is intended to address the complexity and uncertainty of the selection process in location decision problems. That is, the use of DEMATEL methodology to analyze the causal and effect relations among criteria, ANP to provide criteria priorities, and AHP to rank potential collection and distribution locations. Additionally, fuzzy set theory (FST) is employed to deal with the vagueness, ambiguity, and uncertainty of human judgments (Zadeh, 1965 [23]), prevalent in carrying out the three identified MCDM methodologies (i.e., DEMATEL, ANP, and AHP). The three MCDM methodologies, along with FST, are used as they are suitable to address the following conditions of locating collection and distribution centers for reverse logistics. The problem requires a selection of the best location among possible location sites for collection and distribution functions, subject to multiple and often conflicting criteria. The problem seems to be in a simple hierarchical structure (i.e., goal, criteria, and alternatives); however, real-life conditions suggest that the set of criteria contains interrelationships which must be captured to address the problem holistically. These interrelationships, thus, are identified by fuzzy DEMATEL, and the fuzzy ANP approach is used to address them and generate criteria weights. Finally, to identify the best alternative for each function, the fuzzy AHP methodology is used to generate the relative weights of the possible alternative locations. As opposed to other MCDM approaches for this purpose (e.g., TOPSIS, PROMETHEE, ELECTRE, VIKOR, to name a few) where rankings are directly generated, the fuzzy AHP method which produces priority weights of the alternatives provides a meaningful scheme for allowing tradeoffs of a location between collection and distribution functions. Such a tradeoff is instrumental for operational viability. The gap that is advanced in this paper is twofold: (a) the use of evaluation criteria to identify collection and distribution centers in the literature is found to be only a few and limited and is focused on single-objective-based mathematical models; this paper seeks to consider critical elements of logistics infrastructure and its relations among one another to determine collection and distribution centers using an integrated MCDM model, and (b) the concept of remanufac-tured products by consumers in developing countries, such as the Philippines, is relatively unwelcome; this paper pursues to address issues in remanufacturing particularly with that of limited and ill-informed policies, cultural preferences, and assessment of actual benefits. 2. Literature review The collection of used products is one of the most important tasks of reverse logistics and also the first task that affects all the other activities in remanufacturing. EOL product collection is accompanied by uncertainty, especially in terms of quantity that must be taken into consideration in establishing collection centers (Serrano et al., 2013 [24]). On the other hand, distribution of goods involves the transportation of both finished and raw materials; its objective is to make sure that the products delivered are in good condition and will arrive at the right destination. In distribution systems, distribution centers are sometimes required to connect manufacturers and customers for supporting and improving the product flow (Langevin et al., 1996 [25]; Yang, 2013 [26]). Due to uncertainties related to returns (e.g., timing, quality, quantity, disassembly and reassembly, homogeneity of product range), the collection function is challenged (Mukher-jee and Mondal, 2009 [27]). In the same manner, the distribution function can potentially be affected when end users do not support remanufactured goods due to negative user perception and unawareness of its quality and price, to name a few (Choudhary and Singh, 2011 [28]; Serrano et al., 2013 [24]; Sharma et al., 2016 [29]). To address such issues on the collection and distribution process related to remanufacturing, collection and distribution centers are utilized (Malik et al., 2015 [14]; Pop et al., 2015 [15]). These two logistics functions may be performed in a center altogether or separately. In an ideal situation, a collection of EOL products and distribution of remanufactured products must be optimized; that is, the collection rate and sales must be maximized. It is, however, significant to note that the type of collection model likewise affects this situation. For instance, a recent study conducted by Hong and Yeh (2012) [16] compared non-retailer collection model and retailer Advances in Production Engineering & Management 14(3) 2019 299 Ocampo, Himang, Kumar, Brezocnik collection model and found that the latter is superior to the other model under certain aspects such as profit and return rate. Also, when there is a consideration on operating channels involved (i.e., retailer-managed collection, manufacturer-managed collection, and third-party-managed collection), a retailer-managed collection is believed to be optimal when there are economies of scale; otherwise, the manufacturer-managed collection becomes an optimal option (Atasu et al., 2013 [17]). Furthermore, collection points in a reverse logistics system location have also been a focus of relevant studies. Wojanowski et al. (2007) [18] proposed combining a collection of used products with retail activities. A continuous modeling framework is presented for designing a dropoff facility network. They determined that a primary factor for an organization to be involved in the collection of used products is the net value that can be reacquired from a returned product. On the other hand, a mixed-integer nonlinear model is presented by Min and Ko (2008) [19] in determining the optimal number and locations of collection points as well as its centralized return centers. It is proposed to enhance customer convenience by reducing travel time and effort to return used products, thereby, improving the efficiency of product returns. Therefore, an adequate number of collection facilities need to be situated proximate to that of the customers. Similarly, Malik et al. (2015) [14] presented other techniques such as graph theory and matrix approach to determine viable locations for collection centers based on ten key factors, comparative significance, and its availabilities. Other authors have also developed mathematical models for the design of reverse logistics network design, considering the location and allocation of facilities (Mutha and Pokharel, 2009 [1]; Yi et al., 2016 [30]). As for the distribution centers, determining practical locations are considered an essential problem as that of collection centers which have also served as the focal point of studies in re-manufacturing for the past few decades (Owen and Daskin, 1998 [31]). Two problems of the most highly studied problems for facility location are the p-median problem and the maximal covering location problem. The p-median problem concerns on locating p facilities to minimize the total demand-weighted distance between each customer to the nearest facility around. For the maximal covering location problem, its objective is to locate a fixed number of distribution facilities to make sure that the number of covered demands is maximized. The two models share a common objective; that is, to be able to get the attention of customers to maximize revenue (Zhang et al., 2016 [32]). Furthermore, the total relevant cost for the whole distribution process can be minimized when the proper selection of facility location is made (Kuo et al., 2011 [33]). Reverse logistics studies for developing countries are unsurprisingly scarce as it is still in a state of infancy (Sarkis et al., 2010 [34]; Zhang et al., 2011 [35]). In fact, there are still many aspects that need to be considered and explored in the strategic planning of collection centers location. At a broader scope, remanufacturing is popular in developed economies considering its advantages (Sharma et al., 2016 [29]). Developed economies have a more mature foundation on remanufacturing as it is practiced as a means to deal with EOL issues. In developed economies, a well-established understanding and perception of environmental issues exist (Nunes et al., 2009 [36]). Additionally, governments in developed countries implement policies that promote the growth of remanufacturing (Govindan et al., 2016 [37, 38]). Consequently, more research regarding sustainability approaches like reverse logistics has been focused on developed countries (Sarkis et al., 2010 [34]; Zhang et al., 2011 [35]). Consumers in well-developed Western countries are more receptive of remanufactured products, while the opposite situation is observed in most developing countries (Nnorom and Osibanjo, 2008 [12]; Zou et al., 2016 [13]). Poor knowledge, limited consumer acceptance, scarcity of remanufacturing tools and techniques, poor remanufacturability of many products, and quality concerns hinder and significantly limit the potential for developing countries from practicing remanufacturing. OEM practices such as patents and intellectual property rights are also hindrances to remanufacturing as they limit possible remanufacturing operations only to the OEM (Ijomah et al., 2007 [39]). Sustainable development in developing countries is relatively lower, as is evident in some countries like Thailand, Vietnam, India, Malaysia, and the Philippines (Xuetal., 2013 [40]). Complete legislation systems in the context of remanufacturing in these countries are not yet fully developed since there is no recognition of the importance of remanufacturing in most firms in developing countries. Hence, empirical data, specifically in the Philippines, is deficient (Saavedra et al., 2013 [41]). 300 Advances in Production Engineering & Management 14(3) 2019 A novel multiple criteria decision-making approach based on fuzzy DEMATEL, fuzzy ANP and fuzzy AHP for mapping ... 3. Methodology The following subsections present the MCDM methodologies to be integrated into this work for determining a location for collection and distribution centers. 3. 1 Fuzzy set theory The fuzzy set theory was developed to deal with uncertainty and impreciseness of human decision (Zadeh, 1965 [23]). In a set of collection of objects x e X where X is the universe of discourse and A Q X, the classical set theory defines the membership of or x £ A with truth values defined in a membership function in Eq. 1. A is a crisp set if ^A(x):X ^ {0,1}. (i) xeA] L0 x£A\ A is a standard fuzzy set if 3 a membership function ^A(x) such that ^A(x):X ^ [0,1]. The set of 2-tuple A = {x,^A{x):x e X,^A(x) e [0,1]} is a fuzzy set where and ^A(x) is the membership function of x e A. Fuzzy numbers are fuzzy subsets of R. Fuzzy number foundations and their arithmetic operations were first introduced by Zadeh (1965) [23]. Commonly used in fuzzy set theory applications, a fuzzy number is defined as a convex normalized fuzzy set in R with membership function which is piecewise continuous. In MCDM applications, a left-right (L-R) fuzzy number is commonly adopted. A fuzzy number A is of L-R type if 3 membership functions for left and for right with i,rel, and (2) l,r>0 with {R((x-M)/r) x>M) (3) (4) where Mel is the modal value of A and i,re^ are the left and right spreads of A. In this work, an L-R type triangular fuzzy number (TFN) was adopted because of its popularity and ease of implementation (Promentilla et al., 2008 [42]). A triangular fuzzy number expresses the strength of each pair of elements in the same group and can be denoted as A = (l,m,u) (5) where I u (6) where l,m,r e R, ^A(x) ^ [0,1] and X is the universe of discourse. The arithmetic operations of two TFNs denoted by (at,a2,a3) and (b1,b2,b3) are shown in Eqs. 7 to 10. A + B = (a± + bll a2 + b2, a3 + b3) (7) Advances in Production Engineering & Management 14(3) 2019 301 Ocampo, Himang, Kumar, Brezocnik A-B = (a1- b±, a2 - b2, a3 - b3) A 0 B =(a1b1, a2b2, a3b3) A + B =(a1/b3, a2/b2, a3/b±) (8) (9) (10) Linguistic scales may be used to help decision-makers compare criteria or elements. Scales used by Tseng (2011) [43] and Tseng et al. (2008) [44], presented in Tables 2 and 3, were adopted in this study. The linguistic scales are assigned to numbers in a triangular fuzzy scale (i.e., for both Table 2 and Table 3), as well as its reciprocals in the pairwise comparison matrix (i.e., as shown only in Table 3). These tables have an ascending order for the triangular fuzzy numbers along with the degree of importance for each scale. This scale helps address vagueness in decision making by allowing qualitative answers to be quantified. Further, the concept of the fuzzy set theory is integrated into the conventional DEMATEL, ANP, and AHP methodologies to obtain a more comprehensive judgment of decision-makers. 3.2 Fuzzy DEMATEL methodology The DEMATEL methodology roots from the need to enable analyses and solve problems utilizing pragmatic visualization method specifically directed graphs (Gabus and Fontela, 1972, 1973 [45, 46]; Herrera et al., 2000; Wang and Chuu, 2004 [47]; Tsai and Chou, 2009 [48]). These directed graphs, also known as digraphs, are believed to be more useful compared to directionless graphs because digraphs illustrate directed relations (i.e., causal and effect relations) of sub-systems. When directed relations are established well, it can provide a better understanding of system elements in a complex setting. While conventional DEMATEL methodology is proven effective in evaluating factor relations, human judgment on decision variables remains subjective; thus, crisp values become inadequate (Büyüközkan and £iffi, 2012 [49]). Hence, the fuzzy set theory is applied to the conventional DEMATEL methodology. Fuzzy DEMATEL has been widely applied in various areas such as air transportation system (Bongo and Ocampo, 2017 [50]), supplier evaluation problems (Büyüközkan and £iffi, 2012 [49]), green supply chain management practices (Lin, 2013 [51]), truck selection problem (Baykasoglu et al., 2013 [52]), firm environmental knowledge management (Tseng, 2011) [43], and monitoring of paint utilization (Kumar et al., 2017 [53]), to name a few. 3.3 Fuzzy analytic hierarchy process (AHP) and fuzzy analytic network process (ANP) Saaty (1977) [54] developed the analytic hierarchy process (AHP) to simplify complex decision problems by structuring the decision attributes and alternatives in a hierarchical manner using a series of pairwise comparisons. AHP models are best used in a decision problem where there are unidirectional hierarchical relations among levels. When the relationships between levels do not merely signify higher or lower, dominant or subordinate, direct or indirect interactions, the analytic network process (ANP) may be used instead. ANP is also introduced by Saaty (1996) [55] as an extension of AHP where feedback mechanisms in a network type of structure are utilized to illustrate better the dependence among alternatives or attributes by obtaining composite weights through a supermatrix (Shyur, 2006) [56]. Both the traditional AHP and ANP methodology are unable to handle imprecise judgments elicited by decision-makers, thus, the enhancement of such methodologies in the being of fuzzy AHP and fuzzy ANP (Tavana et al., 2013) [57]. Instead of a crisp value used in the evaluation process, fuzzy AHP and fuzzy ANP adopt a range of linguistic expressions with a corresponding triangular fuzzy number to improve how decision-makers make qualitative evaluations. Recent applications of fuzzy AHP are, among others, selection of an R&D strategic alliance partner (Chen et al., 2010 [58]), selection of best pricing strategy for new product development (Liao, 2011) [59], resolution of uncertainty and imprecision in the evaluation of airlines' competitiveness (Wu et al., 2013 [60]), selection of a cruise port of call location (Wang et al., 2014 [61]), selection of passenger aircraft type (Dozic et al., 2017 [62]), and various applications in automotive industry (Banduka et al., 2018 [63]). As for fuzzy ANP, it has been widely applied in areas such as evaluation and selection of suppliers (Razmi et al., 2009 [64]), selection of conceptual design in a product development (Ayag and Özdemir, 2009 [65]), prioritization of strategy (Babaesmailli et al., 2012 [66]), prioritization of advanced-technology projects at the National Aeronautics and Space Administration (NASA) (Tavana et al., 2013 [57]), and evaluation and selection of outsourcing providers for a telecommunication company (Uygun et al., 2014 [67]). 302 Advances in Production Engineering & Management 14(3) 2019 A novel multiple criteria decision-making approach based on fuzzy DEMATEL, fuzzy ANP and fuzzy AHP for mapping ... 3.4 An integrated MCDM methodology A detailed procedure on the integrated MCDM approach to determine collection and distribution centers is shown as follows: Step 1: Apply fuzzy DEMATEL. The interrelationships between criteria are established by decisionmakers using linguistic rating scales adopted from Tseng (2011) [43] (see Table 2). This scale helps address vagueness in the decision-making process by allowing qualitative answers to be quantified. Fuzzy DEMATEL is utilized to identify causal and effect criteria. The fuzzy DEMATEL method includes the following steps (Lin and Wu, 2004 [68]). These steps are further applied to both models (e.g., collection and distribution centers) considered in this paper. Table 2: Linguistic scale for DEMATEL as adopted from Tseng (2011) [43] Linguistic variable Code (TFNs) No influence NI (0.0, 0.1, 0.3) Very low influence VLI (0.1, 0.3, 0.5) Low influence LI (0.3, 0.5, 0.7) High influence HI (0.5, 0.7, 0.9) Very high influence VHI (0.7, 0.9, 1.0) 1.1 Compute for the fuzzy initial direct-relation matrix. The fuzzy initial direct-relation matrix Z involves fuzzy numbers represented as ZtjK = (lijK,mijK,uijK) as shown in Eq. 11 where k represents the kth decision-maker. 0 Zi2 Z\n C^ijk)nxn ■^21 0 • ■■ z2n -^nl Z-n.2 0 (11) 1.2 Obtain the average judgment of decision-makers. The average judgment of k decision-makers also referred to as matrix Z, is obtained using Eq. 12: Z± © © * * * © Zfr z = (12) 1.3 Solve for the fuzzy normalized direct relation matrix X This matrix is obtained using Eq. 13 where X^ = —= and r = max1). According to the authors Lin and Wu (2004) [68], the transformation of linear scale is used as a formula for normalization to transform the criteria scales into its corresponding scales. For instance, at represents each triangular fuzzy number in each cell of Ztj and r on the other hand is the maximum summation of the upper bound element of each TFN in every row of Eq. 14. X = IL / IL IL j=í \j=í j=í ^11^12 ■■ ^21^22 ■■ n ■■ Xnn n ij+YJu ), 7 = 1 / Y ~~ TTXŒ, I- \}=í (13) (14) 1.4 Define crisp matrix for each element of the matrix X. The elements in the matrix X where xi} = (l'ij,m'ij,u'ij) are extracted to obtain three crisp matrices, as presented in Eqs. 15, 16, and 17, respectively. 17' t il I' 12 * ■ /' 1 L In m'n ™'l2 • • m In u'íí u In I' 21 L 22 . y L 2n (15); Xm = ™'2l ™'22 • • m 2n (16); Xu = w'21 "'22 * u 2n (17) L' nl L n2 . r nnJ ™'n2 • • m nn- -u nl u n2 ^ nn- 1.5 Attain the fuzzy total relation matrix f. This matrix is computed using Eq. 18 where the matrix T contains TFNs as in Eq. 19. Advances in Production Engineering & Management 14(3) 2019 303 Ocampo, Himang, Kumar, Brezocnik txl tl2 tin T = ^22 ¿2 n ¿ni t-n.2 tnn- in which tij = {l"i l"ij), [l"ij]= XiXQ- Xl)~1, [m"ij]= Xm x (/- Xm) -i (18) (19) and [u"i}]= Xux(I-Xu)~1 1.6 Defuzzify total relation matrix f. The entries in the total relation matrix T are defuzzified using the center of gravity equation shown in Eq. 20 to obtain matrix T = (t^ . tU = l"ij + 4m"ij + u"ij (20) 1.7 Set threshold value. The negligible effects are filtered out using a threshold value. This value indicates how one factor affects another. The elements in matrix T that exceed the threshold value are considered of significant relations. The threshold level used in this work is determined by decision-makers. The arithmetic mean of the decision-makers' inputs is computed to determine the threshold. 1.8 Classify the nature of criteria. The Di +Ri and Di — Ri of each criterion are calculated where Di and Rt are rows and columns sum of matrix T, respectively. Dt shows the relative importance of the criteria while Dt —R^ demonstrates a causal relationship. A positive value between the difference of Di and Ri denotes that a criterion belongs to the causal group. Conversely, negative value denotes that a criterion belongs to the effect group. 1.9 Construct the impact network relations map. The relationship of one criterion to another is illustrated through a constructed impact relationship map. A scatter graph is created where a criterion's Dt value is the abscissa and Dt — R^ value as the ordinate. Step 2: Apply FANP. The following steps from Ocampo et al. (2015) [69] below are adapted to generate the criteria weights. These steps are further applied to both models (e.g., collection and distribution centers) considered in this paper. 2.1 Attain initial matrix Ak. The elicited judgment of decision-makers on each criterion based on pairwise comparison is gathered using the linguistic scale with TFNs presented in Table 3. Table 3: FANP linguistic scale from Tseng et al., (2008) [44] Linguistic Scale Code Triangular fuzzy scale Triangular fuzzy reciprocal scale Just equal JU (1,1,1) (1,1,1) Equal importance EQ (1/2,1,3/2) (2/3,1,2) Moderate importance MO (5/2,3,7/2) (2/7,1/3,2/5) Strong importance ST (9/2,5,11/2) (2/11,1/5,2/9) Demonstrated importance DE (13/2,7,15/2) (2/15,1/7,2/13) Extreme importance EX (17/2,9,9) (1/9,1/9,2/17) The initial decision per comparison matrix ¿lfeis equivalent to (ai]k)nxn represented as a-ijk = (.hjk,mijk,uijk) where k represents the kth decision-maker. The form of this matrix is shown in Eq. 21: <1,1,1) a12 a21 (1,1,1) aln a2n iP-ijk)nxn ~ : : •. : (21) anl an2 ••• (1,1,1). 2.2 Aggregate the judgments using the geometric mean method. The judgment of the decisionmakers elicited for each matrix type is then aggregated. The geometric mean method is among the most commonly used methods for aggregating individual ratings (Saaty, 2008) [70]. This method generates an aggregate fuzzy pairwise comparison matrix A = (lij,mij,uij) , shown in Eq. 22, is used in this paper. (22) 304 Advances in Production Engineering & Management 14(3) 2019 A novel multiple criteria decision-making approach based on fuzzy DEMATEL, fuzzy ANP and fuzzy AHP for mapping ... where ltj, m^, and u^- represents lower bound, modal value, and upper bound, of the aggregate fuzzy judgment, respectively. 2.3 Compute for consistency ratio of each pairwise comparison. Using this approach, the consistency of the initial set of fuzzy judgments made by decision-makers is measured (see Eq. 23). If the optimal value (2) is positive, all solution ratios completely satisfy the fuzzy judgments, which mean that the initial set of fuzzy judgments is consistent. On the other hand, if the optimal value (2) is negative, the solution ratios of the fuzzy judgments are strongly inconsistent. Max A subject to: (™-ii -Wi + lijWj <0 (uij -mij)Awj +wi-uijwj <0 ,„3, Yn ™ -1 ( ) Lk=lwk - 1 wk > 0, k = 1,2, ...,n i = 1,2, ...,n — 1,j = 2,3, ...j >i where A represents the degree of satisfaction of fuzzy constraints, is lower bound of the TFN in each element, m^ is the modal value of the TFN in each element, utj represents the upper bound of the TFN in each element, W; represents the crisp weight of row element, and Wj is the crisp weight of column element. Eq. 23 is run in LINGO® Optimization Software, where A represents the degree of satisfaction of fuzzy constraints, and W; and Wj are the weights of the elements in the pairwise comparison matrix. This formula is used to defuzzify matrices and to give weight to the criteria compared. Each cell in the pairwise comparison is subjected to the constraints and is added in the formula. As suggested by Mikhailov and Tsvetinov (2004) [71], some cells could be removed in case of inconsistency. Some matrices in this study contain only n — 1 cells, the minimum number of cells needed, where n is the number of objects compared. 2.4 Structure of the initial supermatrix. Using the local weights obtained in the previous step, the supermatrix is structured as in Eq. 24. /s11 s12 S21 S22 (24) 2.5 Normalize initial supermatrix. The supermatrix is normalized to achieve a stochastic column matrix by utilizing the column sum (Eq. 25) and dividing each element in the column by Eq. 26 where C is the column sum. T = lim (X1 + X2 + ■■• + Xk) (25) k^rn v- J C = Li=l 2.6 Limit matrix to large powers. The final weights are obtained by raising the matrix to large powers until the elements in the normalized matrix have reached a steady state. 2.7 Determine normalized weights. The criterion interactions are divided by the sum of the interactions of the criteria in an arbitrarily chosen column to determine its corresponding normalized weights. The weights of criteria are used to construct the matrix wc and matrix wd for collection and distribution function, respectively. Step 3: Apply FAHP. The weights of each location in terms of its viability as a collection and distribution center with respect to a function's criteria are determined through FAHP. The steps below are adapted from Wang and Chin (2011) [72] and are applied to both models of this paper. 3.1 Decision-makers elicit their judgment on a pairwise comparison matrix among location alternatives with respect to each criterion. 3.2 The results from the pairwise comparison are then utilized and further follow steps 2.1 to 2.3, accordingly. Advances in Production Engineering & Management 14(3) 2019 305 Ocampo, Himang, Kumar, Brezocnik 3.3 The computed weights for each alternative with respect to a certain criterion are plugged into the matrix Wd. /wdll wdl2 ••• wdlnv / Wd21 Wd22 - Wd2n \ Wd = 1 | ;22 , ;2n I (27] Wi wdn2 ••• wdnn/ 3.4 The global weights for each alternative are determined by multiplying the matrices in steps 2.7 and 3.3. 3.5 The alternatives are ranked according to the global weights identified in the model. A higher global weight denotes a higher priority. Step 4: Obtain a satisfactory map. This map which plots alternatives with its global weights for both functions as its coordinates. This graph allows a comparison between a location alternative's satisfaction for collection and distribution. The satisfaction level represents a location's capability to carry out a function. 4. A case study This section highlights the application of an integrated MCDM approach in the context of identifying the locations for collection and distribution functions of reverse logistics (see Fig. 2]. The decision models are tested in a furniture firm as a case study in Cebu, the Philippines, since this industry produce highly remanufacturable products. Moreover, Cebu is considered as one of the emerging industrialized regions in the Philippines that practice remanufacturing. The MCDM procedure begins with the definition of the decision goal, which is the selection of a viable collection and distribution center location. Then, the location criteria applicable in the Philippine setting for collection and distribution centers are determined through a preliminary survey conducted among experts. A criterion is generally perceived applicable when at least 65 % of the experts agree to have it included in the final roster to be used in the MCDM method (Krishnan and Poulose, 2016 [73]]. Applicable criteria are then included in the second level of the framework for collection and distribution centers presented in Figs. 2 and 3, respectively. The next step involves the implementation of fuzzy DEMATEL to determine the interrelationships among criteria. Afterward, FANP is used to obtain the weights of each criterion. Lastly, possible collection and distribution location points are evaluated using FAHP, which results in a final ranking of alternatives. For further analyses and visual purposes, a two-way graph is used to plot the locations points of collection and distribution centers. Defining the decision goal Identifying criteria for collection and distribution locations Applying Fuzzy DEMATEL to determine interrelationship <> Applying FANP to determine the weight of each criterion <> Identifying possible collection and distribution locations points Applying FAHP to determine the priorities of each alternative Ranking final alternatives <> Plotting results in a two-way graph with respect to collection and distribution location points 306 Fig. 2 Computational framework of the study Advances in Production Engineering & Management 14(3) 2019 A novel multiple criteria decision-making approach based on fuzzy DEMATEL, fuzzy ANP and fuzzy AHP for mapping .. Selection of experts It is crucial that the decision-makers involved in any MCDM problem are carefully selected as the accuracy of evaluation among criteria and alternatives significantly relies on their expertise. In this paper, 10 decision-makers (i.e., two decision-makers from the manufacturing firms, two decision-makers from logistics industries, two academics having research interest in the related fields, two decision-makers from government agencies, and two critical consumers) are tapped. Distinct qualifications are set for each decision-maker ranging from minimum educational attainment, years of experience in related fields, working knowledge in remanufacturing/forward and reverse logistics issues, to knowledge in government legislation for manufacturing industries. The choice of respondents is consistent with that of outstanding MCDM applications conducted by several authors (Govindan et al., 2009 [74, 75]; Mittal and Sangwan 2014 [76]; and Ocampo and Promentilla 2016 [77]) in various areas of concern. Decision models The proposed decision-making models of this paper that pertains to collection and distribution centers are presented in Fig. 3 and Fig. 4, respectively. A hierarchical structure is utilized to solve the location problem. The structure consists of three levels, and each level represents a particular aspect. The goal of this paper is to determine collection and distribution centers as represented by the first level of the structure. The secondary levels are composed of critical criteria for collection and distribution locations. These criteria are believed to have an interrelationship among one another; thus, fuzzy DEMATEL and FANP are applied in the secondary levels to evaluate the interrelationships and to generate corresponding weights. The location points, as alternatives, are represented in the third level. FAHP is used at this level to determine priorities in terms of an alternative's ranking as a collection or distribution center. The proposed framework is applied through a case study in the furniture industry. Legend: CC - collection center Cn - criteria for collection center Ln - location alternatives Fig. 3 The decision model for the collection center Fig. 4 The decision model for the distribution center Advances in Production Engineering & Management 14(3) 2019 Legend: DC - distribution center Dn - criteria for distribution center Ln - location alternatives 307 Ocampo, Himang, Kumar, Brezocnik Decision criteria In the Philippine setting, nine criteria are agreed by decision-makers from the preliminary survey to apply to the decision problem that concerns the selection of locations for collection and distribution centers. These criteria, coded as C1 through C9 for collection center criteria and D1 through D9 for distribution center criteria, are summarized in Tables 4 and 5. Table 4 Critical criteria for the collection center Code Criteria Description C1 The capacity of the center The holding capacity of the facility. C2 Initial investment The capability of shareholders' financial support in setting up the facility. C3 Government policies and regulations The compliance of requirements given by the government (local regulations on zoning, building codes, among others). C4 Environmental collaboration with The market's interest and acceptance of the remanufactured product for customers environmental preservation. C5 Material availability The availability of end-of-life products in an area. C6 Proper disposal The effective disposal of waste from the facility without any public disturbance. C7 Land price The value of the land per square meter. C8 Supply of product return The number of EOL products that can be collected. C9 Quality of product return The quality of EOL product collected. Table 5 Critical criteria for the distribution center Code Criteria Description D1 Distance from facility between The proximity of competition in a nearby area. competitor D2 The demand for the second market of the The adaptation and acceptance of the remanufactured product by the area secondary market in an area. D3 Initial investment The capability of shareholders' financial support in setting up the facility. D4 Government policies and regulations The compliance of requirements given by the government (local regulations on zoning, building codes, among others). D5 Environmental collaboration with The market's interest and acceptance of the remanufactured product for customers environmental preservation. D6 Distance to suppliers The accessibility and proximity of facility location from suppliers. D7 Transportation The transport of materials and products to and from the location. D8 Proximity to customers The proximity of potential customers of the area. D9 Land price The value of the land per square meter. A more concrete illustration of each criterion about its role in the collection and distribution functions of remanufacturing is given as follows: • The capacity of the center (C1). The holding capacity of a facility is an essential factor in setting up a center. This affects the ability of the center to execute its function. For instance, if the facility has reached its maximum capacity, then it is difficult to store additional units. • Initial investment (C2 and D3). Investment cost in setting up a facility is a factor to consider in the establishment of both a collection and distribution center. Once a facility is established, it is challenging and costly to revert. Initial investment (C2 and D3) covers the cost for construction, labor, materials, and other activities except for land acquisition. • Government policies and regulations (C3 and D4). Government legislation has been identified by Sharma et al. (2016) [29] as an important factor in adopting remanufacturing in a developing country. A government's support regarding remanufacturing can either be a major driver for remanufacturing (Xiang and Ming, 2011 [78]), or a major roadblock (Sharma et al., 2016 [29]). • Environmental collaboration with customers (C4 and D5). Environmental collaboration with customers is achieved when there is support for sustainable practices. The level of acceptance of remanufacturing and support for sustainable practice is directly proportional; when acceptance is high, support is also high (Andel and Aichlmayr, 2002 [79]). This, in turn, creates greater collaboration with the public especially the customers. Cooperation of customers in a distribution function concerns with the support of remanufac-tured products. • Material availability (C5). The ability to collect is vital in remanufacturing since the raw materials are used products. A lack and insufficient amount of EOL products may hinder the remanufacturing operations as a collection of the used products is the first step that affects all other activities in remanufacturing. With this, material availability (C5) is a significant consideration to assess the viability of a collection center. 308 Advances in Production Engineering & Management 14(3) 2019 A novel multiple criteria decision-making approach based on fuzzy DEMATEL, fuzzy ANP and fuzzy AHP for mapping ... • Proper disposal (C6). It is necessary to protect the natural environment, and reduce pollution caused by pre-remanufacturing activities. If there is improper waste disposal for facilities, the surrounding area takes a negative impact that affects the community (McAllister, 2015 [80]). • Land price (C7 and D9). Similar to investment cost, the price of acquiring land is an essential consideration since it is the foundation for establishing a collection and distribution center. • Supply of product return (C8). Supply of product return is an important criterion since some EOL units that can be collected greatly affects the input cost of materials and components. For instance, organizations would have to acquire new components if there are limited EOL units collected. • Quality of product return (C9). Quality uncertainty must be addressed similarly with quantity uncertainty in which a location with a higher quality level of collected units is preferred. The quality of the EOL units to be collected in an area is also an important consideration since it affects the remanufacturing suitability of the components. • The distance of facility between competitor (D1). It is important to know the distance between competitors to assess how competitive the area is. This enables the organization to know if they can conduct business, perform operations, and penetrate the market. • The demand for the secondary market of the area (D2). The adaptation and acceptance of remanufactured products by the secondary market in an area affect the distribution function of remanufactured goods. For instance, if the end consumers are not supportive of remanufactured goods, then the ability to distribute is negatively affected (Choudhary and Singh, 2011 [28]). As some customers are hesitant in accepting remanufactured products due to its negative perception on its quality (Sharma et al., 2016 [29]), it is of great importance to ensure the consumers' needs are fulfilled upon creating a network supporting the distribution center. • Distance to suppliers (D6). The accessibility and proximity of facility location from suppliers is a significant consideration that affects the lead time of acquiring supplies as it may affect the efficiency of operations if supplies are low. • Transportation (D7). The distribution of remanufactured products as part of the reverse logistics practice involves the transportation of remanufactured products from one location to the other. Transportation (D7) pertains to the ease of product movement, the available alternative routes, and available mode of transportation. • Proximity to customers (D8). The proximity to customers is an important criterion in selecting a location for a distribution center. It is one of the primary considerations as it affects economic performance. It can be observed that there are criteria applicable for both collection and distribution centers. Some criteria are exclusive to a specific facility. Examples of the limited list of criteria are material availability (C5) for collection centers and proximity to customers (D8) for distribution centers. These criteria are exclusive to a particular facility as they are essential considerations to meet specific objectives. Material availability (C5) is deemed crucial by the decisionmakers as applicable only for collection centers as the facility's primary focus is to acquire EOL units. While in the distribution function, this criterion is irrelevant since distribution centers do not deal with EOL unit acquisition. Proximity to customers (D8) is deemed as applicable only for distribution centers since a facility located in the vicinity of customers would significantly minimize costs for delivering products to destinations. 5. Empirical results of the case study Firstly, the fuzzy DEMATEL methodology is carried out. The aggregate direct relation matrices of the decision-makers are computed using Eq. 12 and are shown in Tables 6 and 7. It is then normalized using Eqs. 13 and 14 which results are presented in Tables 8 and 9. Note that the normalized direct relation matrices are still expressed in fuzzy numbers, therefore, Eq. 20 is used to Advances in Production Engineering & Management 14(3) 2019 309 Ocampo, Himang, Kumar, Brezocnik defuzzify the values and obtain the total defuzzified relation matrices as in Tables 10 and 11. These total defuzzified relation matrices are evaluated by decision-makers further as to which relations are perceived to be significant. The arithmetic average of decision-makers' inputs regarding significant relations among criteria represents the threshold value set. For the case of this paper, a threshold value of 0.47 is established. Then, the next step involves the identification of relations among criteria. Tables 12 and 13 shows the influence and effect of the criteria. The term (D + R) indicates the relative importance of a criterion while (D — R) determines whether a criterion is a dispatcher (net cause) or a receiver (net effect). When a criterion has a positive (D — R) value, it implies that it influences other criteria; otherwise, it is the one influenced. As can be noted from the results, government policies and regulations (C3), environmental collaboration with customers (C4), material availability (C5), land price (C7), supply of product return (C8), and quality of product return (C9) influences the two remaining criteria for collection center whereas demand of the second market of the area (D2), government policies and regulations (D4), environmental collaboration with customers (D5), distance to suppliers (D6), transportation (D7), proximity to customers (D8), and land price (D9) influence the remaining critical criteria for distribution center. On the other hand, Figs. 5 and 6 show the interdependent relationships of criterion i to criterion j for a collection and distribution center, respectively. The criteria are plotted in a scatter graph where D^ +Rt is its abscissa and Di —Ri its ordinate. The elements of the deffuzified matrices are compared to the threshold value set. A one and zero representation are developed to distinguish the significant relationship between criteria. A value of one represents a significant relationship between criterion i to criterion j, while a value of zero means no significance between criteria. The arrows denote the influence given and received by one criterion to the other. The arrowhead represents the criteria being affected while the tail corresponds to the influencing criterion. It can also be noted in Fig. 5 that government policies and regulations (C3), material availability (C5), and of product return (C8) have mostly affected other criteria. However, government policies and regulations (C3) is not affected by any criteria, while material availability (C5) and product return (C8) affect each other. The capacity of the center (C1) and initial investment (C2) are mostly affected by other criteria except for the quality of product return (C9), thus, indicates its dependence on other criteria. In Fig. 6, proximity to customers (D8) has mostly affected other criteria and is affected by the demand of the second market of the area (D2) and environmental collaboration with customers (D5). The demand of the second market of the area (D2) is mostly affected by the other criteria namely: distance of facility between competitor (D1), environmental collaboration with customers (D5), proximity to customers (D8) and land price (D9). The demand for the second market of the area (D2) is mainly dependent on other criteria. Once the evaluation of criteria using fuzzy DEMATEL approach is completed, FANP and FAHP are correspondingly implemented. These methodologies focus on comparing critical criteria with its significance in a collection and distribution center and identifying interrelationships among criteria. The first step involves aggregating the elicited judgment of decision-makers using Eq. 22. Then, the consistency of each matrix is computed using LINGO® software following through Eq. 23. A positive value of A indicates that an aggregate matrix has acceptable consistency; conversely, a negative value indicates an inconsistent matrix. In cases of inconsistencies, cells can be deleted (Mikhailov and Tsvetinov, 2004 [71]). Due to some inconsistencies in the judgment of decision-makers, only the first row, being n — 1, is considered since n — 1 is the minimum solution required in LINGO® to solve Eq. 23. The initial supermatrices in Tables 14 and 15 are constructed using the generated weights from Eq. 22 and are normalized using Eqs. 25 and 26. Tables 16 and 17 show the normalized matrices for collection and distribution function, respectively. The final weights are obtained by raising these normalized matrices into large powers until a steady-state behavior is observed. The final weights listed in Table 18 are representative of the matrix wc and matrix wd of collection and distribution functions, respectively. The furniture firm considered in this paper provided four location alternatives under evaluation. Table 19 summarizes the details of these location alternatives, including lot area, land price, and zoning, to name a few. The location alternatives 310 Advances in Production Engineering & Management 14(3) 2019 A novel multiple criteria decision-making approach based on fuzzy DEMATEL, fuzzy ANP and fuzzy AHP for mapping ... are evaluated by decision-makers to determine its viability in terms of priorities with respect to each criterion for collection center and distribution center. The elicited judgment of decisionmakers is aggregated using the geometric mean method as shown in Eq. 22. Then, using LINGO®, consistency ratios, as well as local weights of the alternative locations, are computed. The product of matrix wc and wd representing the final weights of each criterion (see Table 18) and matrix WfC and Wfd (see Tables 20 and 21) is computed to generate the global weights of each alternative (see Table 22). In reference to the global weights, a scatter graph is constructed, as shown in Fig. 7 to map the location alternatives' satisfaction being a collection and distribution center. The satisfaction level represents a location's capability to carry out a function. Table 6 Aggregated direct relation matrix for the collection center CI (0.0, 0.1, 0.3) (0.5, 0 7, 0.9) (0.3, 0.5, 0.7) (0.4, 0 S, 0.8) (0.4, 0 5, 0.7) (0.4, 0.6 0.8) (0.4, 0.6, 0.8) (0.4, 0 6, 0.7) (0.3, 0.4, 0.6) C2 (0.6, 0.8, 0.9) (0.0, 0 1, 0.3) (0.4, 0.6, 0.8) (0.4, 0 6, 0.8) (0.4, 0 6, 0.7) (0.3, 0.5 0.7) (0.5, 0.7, 0.9) (0.3, 0 5, 0.7) (0.2, 0.3, 0.5) C3 (0.4, 0.6, 0.8) (0-4, 0 6, 0.8) (0.0, 0.1, 0.3) (0.5, 0 7, 0.9) (0.4, 0 5, 0.7) (0.6, 0.8 0.9) (0.6, 0.8, 0.9) (0.3, 0 5, 0.7) (0.3, 0.4, 0.6) C4 (0.5, 0.7, 0.8) (0.4, 0 6, 0.7) (0.5, 0.7, 0.9) (0.0, 0 1, 0.3) (0.4, 0 6, 0.8) (0.5, 0.7 0.9) (0.3. 0.4, 0.6) (0.5, 0 6, 0.8) (0.5, 0.6, 0.8) C5 (0.6, 0.8, 0.9) (0.5, 0 7, 0.9) (0.3, 0.5, 0.7) (0.4, 0 5, 0.7) (0.0, 0 1, 0.3) (0.3, 0.5 0.7) (0.4, 0.6, 0.8) (0.5, 0 7, 0.9) (0.5, 0.7, 0.8) C6 (0.4, 0.6, 0.8) (0.4, 0 6, 0.8) (0.5, 0.7, 0.9) (0.4, 0 6, 0.8) (0.3, 0 5, 0.7) (0.0, 0.1 0.3) (0.4, 0.6, 0.8) (0.3, 0 5, 0.7) (0.3, 0.4, 0.6) C7 (0.6, 0.8, 0.9) (0.5, 0 7, 0.9) (0.4, 0.6, 0.8) (0.4, 0 6, 0.8) (0.4, 0 6, 0.8) (0.4, 0.6 0.8) (0.0, 0.1, 0.3) (0.3, 0 5, 0.7) (0.2, 0.3, 0.5) C8 (0.6, 0.8, 0.9) (0.6, 0 7, 0.9) (0.3, 0.5, 0.7) (0.5, 0 7, 0.8) (0.5, 0 7, 0.9) (0.5, 0.7 0.9) (0.3, 0.5, 0.7) (0.0, 0 1, 0.3) (0.5, 0.7, 0.9) C9 (0.4, 0.5, 0.7) (0.4, 0 5, 0.7) (0.3, 0.5, 0.7) (0.4, 0 6, 0.8) (0.4, 0 6, 0.8) (0.4, 0.6 0.7) (0.2, 0.4, 0.6) (0.4, 0 6, 0.8) (0.0, 0.1, 0.3) Table 7 Aggregated direct relation matrix for distribution center D1 (0.0, 0.1, 0.3) (0.5 0.7 0.9) (0.3, 0.5, 0.7) (0.1, 0.3, 0.5) (0.2, 0.4, 0.6) (0.4, 0.6 0.8) (0.4, 0.6, 0.8) (0.5, 0.7, 0.9) (0.4, 0.6, 0.8) D2 (0.5, 0.7, 0.9) (0.0 0.1 0.3) (0.5, 0.7, 0.9) (0.3, 0.5, 0.7) (0.5, 0.7, 0.9) (0.4, 0.6 0.7) (0.4, 0.6, 0.8) (0.6, 0.7, 0.9) (0.4, 0.6, 0.8) D3 (0.3, 0.4, 0.6) (0.4 0.6 0.8) (0.0, 0.1, 0.3) (0.3 0.5, 0.7) (0.4, 0.6, 0.8) (0.3, 0.4 0.6) (0.4, 0.5, 0.7) (0.4, 0.5, 0.7) (0.5, 0.7, 0.9) D4 (0.2, 0.4, 0.6) (0.4 0.6 0.8) (0.4, 0.6, 0.8) (0.0, 0.1 0.3) (0.5, 0.7, 0.9) (0.1, 0.3 0.5) (0.3 0.5, 0.7) (0.3, 0.5, 0.7) (0.5, 0.7, 0.9) D5 (0.4, 0.6, 0.8) (0.5 0.7 0.9) (0.4, 0.6, 0.8) (0.4, 0.6, 0.8) (0.0, 0.1 0.3) (0.1, 0.3 0.5) (0.2 0.4, 0.6) (0.5, 0.7, 0.9) (0.3, 0.5, 0.7) D6 (0.2, 0.4, 0.6) (0.3 0.5 0.7) (0.4, 0.6, 0.8) (0.1, 0.3 0.5) (0.3 0.4, 0.6) (0.0, 0.1 0.3) (0.5, 0.6, 0.8) (0.4, 0.6, 0.8) (0.3, 0.5, 0.7) D7 (0.5, 0.7, 0.8) (0.4 0.6 0.8) (0.4, 0.6, 0.8) (0.3 0.5, 0.7) (0.3 0.4, 0.6) (0.5, 0.7 0.9) (0.0, 0.1, 0.3) (0.4, 0.6, 0.8) (0.3, 0.4, 0.6) D8 (0.5, 0.7, 0.9) (0.6 0.8 1.0) (0.5, 0.7, 0.9) (0.4, 0.6, 0.8) (0.5, 0.7, 0.9) (0.3, 0.4 0.6) (0.4, 0.6, 0.8) (0.0, 0.1, 0.3) (0.4, 0.6, 0.8) D9 (0.6, 0.8, 0.9) (0.4 0.6 0.7) (0.6, 0.8, 0.9) (0.4, 0.6, 0.8) (0.4, 0.6, 0.8) (0.4, 0.6 0.8) (0.4, 0.6, 0.8) (0.3, 0.5, 0.7) (0.0, 0.1, 0.3) Table 8 Normalized direct relation matrix for collection center CI C2 C3 C4 C5 C6 C7 C8 C9 (0.000, 0.014, 0.043) (0.081, 0.110, 0.134) (0.056, 0.084, 0.111) (0.068, 0.095, 0.118) (0.084, 0.113, 0.136) (0.062, 0.090, 0.116) (0.081, 0.110, 0.132) (0.084, 0.113, 0.134) (0.051, 0.078, 0.105) (0.069, 0.098, i (0.000, 0.014, i (0.061, 0.090, i (0.053, 0.081, i (0.071, 0.098, i (0.061, 0.090, i (0.078, 0.107, i (0.079, 0.107, i (0.052, 0.078, i .126) (0.045, .043) (0.059, .117) (0.000, .107) (0.069, .124) (0.043, .117) (0.071, .132) (0.064, .129) (0.040, .105) (0.040, 0.072, 0.098) 0.087, 0.113) 0.014, 0.043) 0.098, 0.124) 0.069, 0.098) 0.098, 0.123) 0.092, 0.118) 0.066, 0.094) 0.066, 0.095) (0.056, (0.056, (0.072, (0.000, (0.051, (0.058, (0.056, (0.071, (0.061, 0.084, 0.110) 0.084, 0.110) 0.101, 0.126) 0.014, 0.043) 0.078, 0.104) 0.087, 0.113) 0.084, 0.110) 0.098,0.121) 0.087,0.111) (0.051, 0.078, (0.053, 0.081, (0.051, 0.078, (0.061, 0.090, (0.000, 0.014, (0.046, 0.072, (0.058, 0.087, (0.078, 0.107, (0.062, 0.090, 0.104) 0.107) 0.105) 0.114) 0.043) 0.100) 0.114) 0.130) 0.114) (0.055, 0.081, i (0.046, 0.072, i (0.084, 0.113, i (0.075, 0.104, i (0.048, 0.075, i (0.000, 0.014, i (0.058, 0.087, i (0.069, 0.098, i (0.055, 0.081, i .108) (0.059,0.087, .101) (0.077,0.104, .134) (0.081, 0.110, .130) (0.036,0.061, .104) (0.058, 0.084, .043) (0.059,0.087, .114) (0.000,0.014, .124) (0.048,0.072, .105) (0.029,0.052, 0.111) 0.126) 0.134) 0.088) 0.110) 0.111) 0.043) 0.098) 0.081) (0.053, (0.042, (0.043, (0.065, (0.078, (0.043, (0.043, (0.000, (0.055, 0.081, 0.107) 0.069, 0.097) 0.069, 0.097) 0.092, 0.116) 0.107, 0.130) 0.069, 0.098) 0.069, 0.098) 0.014, 0.043) 0.084, 0.111) (0.036, (0.026, (0.036, (0.065, (0.066, (0.036, (0.023, (0.072, (0.000, 0.061,0.088) 0.049, 0.078) 0.061,0.088) 0.092,0.117) 0.095,0.121) 0.058,0.085) 0.046,0.075) 0.101,0.126) 0.014,0.043) Table 9 Normalized direct relation matrix for distribution center D1 D2 D3 D4 D5 D6 D7 D8 D9 (0.000, 0.015, 0.044) (0.080, 0.109, 0.134) (0.040, 0.065, 0.094) (0.032, 0.059, 0.088) (0.055, 0.083, 0.111) (0.035, 0.062, 0.088) (0.068, 0.097, 0.124) (0.077, 0.106, 0.130) (0.083, 0.112, 0.137) (0.080, (0.000, (0.059, (0.055, (0.080, (0.050, (0.062, (0.091, (0.056, 0.109, 0.015, 0.086, 0.083, 0.109, 0.077, 0.088, 0.121, 0.083, 0.133) 0.044) 0.111) 0.111) 0.133) 0.103) 0.114) 0.140) 0.108) (0.044, (0.075, (0.000, (0.065, (0.060, (0.058, (0.063, (0.075, (0.088, 0.071, 0.099) 0.103, 0.127) 0.015, 0.044) 0.094, 0.121) 0.088, 0.117) 0.086, 0.112) 0.091, 0.117) 0.103, 0.127) 0.118, 0.137) (0.019, (0.047, (0.047, (0.000, (0.062, (0.021, (0.043, (0.055, (0.060, 0.044, 0.074) 0.074, 0.102) 0.074, 0.100) 0.015, 0.044) 0.091, 0.119) 0.044, 0.074) 0.071, 0.099) 0.083, 0.112) 0.088, 0.114) (0.029, 0.053, i (0.075, 0.103, i (0.058, 0.086, i (0.071, 0.100, i (0.000, 0.015, i (0.037, 0.062, i (0.040, 0.065, i (0.074, 0.103, i (0.058, 0.086, i .127) .112) .127) .044) .091) .093) .128) .112) (0.063, (0.055, (0.041, (0.021, (0.021, (0.000, (0.071, (0.038, (0.056, 0.091, 0.118) 0.083, 0.108) 0.065, 0.091) 0.044, 0.074) 0.044, 0.074) 0.015, 0.044) 0.100, 0.125) 0.065, 0.091) 0.086, 0.112) (0.063, (0.065, (0.052, (0.041, (0.028, (0.066, (0.000, (0.060, (0.059, 0.091, 0.117) 0.091, 0.1170 0.077, 0.103) 0.068, 0.097) 0.053, 0.083) 0.094, 0.119) 0.015, 0.044) 0.088, 0.114) 0.088, 0.117) (0.074, (0.081, (0.055, (0.047, (0.080, (0.059, (0.060, (0.000, (0.047, 0.103, 0.127) 0.109, 0.130) 0.080, 0.103) 0.074, 0.102) 0.109, 0.131) 0.088, 0.115) 0.088, 0.115) 0.015, 0.044) 0.074, 0.100) (0.062, (0.056, (0.074, (0.071, (0.050, (0.044, (0.038, (0.059, (0.000, 0.091, 0.118) 0.083, 0.111) 0.103, 0.130) 0.100, 0.125) 0.077, 0.105) 0.074, 0.103) 0.065, 0.094) 0.086,0.114) 0.015 0.044) Table 10 Total defuzzified direct-relation matrix for collection center CI C2 C3 C4 C5 C6 C7 C8 C9 CI 0.41671423 0.476950149 0.416258116 0.44310564 0.43035511 0.44635719 0.42995474 0.41827428 0.36865124 C2 0.50792484 0.408431607 0.434668408 0.449010277 0.43837842 0.4452388 0.45097947 0.41379572 0.3632351 C3 0.50826854 0.497324253 0.39010564 0.484384238 0.45604524 0.50026838 0.47548591 0.43301991 0.39037429 C4 0.51907261 0.491339845 0.465257725 0.409352809 0.46776876 0.49523275 0.43455459 0.45492889 0.42085009 C5 0.53938017 0.511456226 0.444103289 0.470638663 0.40520224 0.4739281 0.45798685 0.47135091 0.42673748 Cb 0.48823607 0.473559814 0.443110971 0.449782541 0.42936429 0.39072474 0.43444987 0.41230093 0.36952385 C7 0.52006556 0.502564809 0.450510473 0.460443193 0.45435602 0.46865441 0.38283735 0.42484289 0.37027334 C8 0.55505298 0.533306792 0.455706833 0.502189978 0.50114104 0.50804831 0.46186805 0.40328363 0.44409374 C9 0.46221839 0.448016535 0.401069165 0.435181448 0.4302822 0.43530698 0.38952752 0.41186065 0.31869023 Table 11 Total defuzzified direct-relation matrix for distribution center Dl_D2_D3_D4_D5_D6_D7_D8_D9 D1 0.379387623 0.480950052 0.449107005 0.355974857 0.398427846 0.401252967 0.429624918 0.464235289 0.43958904 D2 0.499068063 0.43753611 0.515212253 0.414225485 0.476585747 0.423420219 0.463501764 0.507449492 0.46988036 D3 0.414977178 0.450323629 0.387785617 0.374624169 0.417376406 0.367750928 0.406175131 0.433553207 0.44110393 D4 0.407038664 0.445713958 0.456690076 0.320107297 0.42869193 0.346980264 0.395642264 0.426458008 0.43661605 D5 0.441128257 0.483235613 0.465869892 0.399854848 0.365684239 0.358575917 0.396311756 0.470506148 0.43032409 D6 0.393166800 0.423333127 0.431411076 0.332420986 0.378872918 0.306933531 0.404514398 0.422381271 0.39687451 D7 0.453769553 0.46626574 0.468360931 0.380607821 0.410538757 0.409776064 0.362695127 0.454595803 0.42049414 D8 0.497746937 0.531523108 0.516629772 0.423355486 0.478309318 0.409164549 0.461531792 0.425390001 0.47374084 D9 0.490979166 0.487382806 0.516854466 0.41782285 0.451853632 0.417988269 0.451893605 0.467194462 0.39939782 Advances in Production Engineering & Management 14(3) 2019 311 Ocampo, Himang, Kumar, Brezocnik Table 12 Relative critical criteria for importance and causal relationship of collection center Table 13 Relative importance and causal relationship of critical criteria for distribution center D+R D-R D+R D-R Cl 3.84662089 4.51693339 8.363554279 -0.6703125 D1 3.798549593 3.977262239 7.775811833 -0.178712646 C2 3.91166264 4.342950029 8.254612673 -0.431287385 D2 4.206879492 4.206264144 8.413143636 0.000615348 C3 4.1352764 3.90079062 8.036067022 0.234485783 D3 3.693670197 4.207921088 7.901591285 -0.514250891 C4 4.15835806 4.104088986 8.262447048 0.054269077 D4 3.663938509 3.418993798 7.082932307 0.244944711 CS 4.20078393 4.012893321 8.213677246 0.187890604 D5 3.811490765 3.806340792 7.617831557 0.005149972 C6 3.89105308 4.163759646 8.054812727 -0.272706566 D6 3.489908617 3.441842708 6.931751325 0.04806591 C7 4.03454804 3.917644347 7.952192383 0.116903689 D7 3.827103931 3.771890756 7.598994687 0.055213175 C8 4.36469135 3.843657816 8.208349169 0.521033538 D8 4.217391799 4.071763681 8.289155479 0.145628118 C9 3.73215312 3.47242936 7.204582479 0.259723759 D9 4.101367077 3.908020774 8.009387851 0.193346303 0.3 a? o.i o -Q.2 -0.3 0.4 Ob 0.6 Fig. 5 Impact relationship map for the collection center ♦ D4 D< R Fig. 6 Impact relationship map for the distribution center Table 14: Initial supermatrix for collection center Goal Cl C2 C3 C4 C5 C6 C7 C8 C9 Goal 1 1 1 1 1 1 1 1 1 Cl 0.051692 0.1657246 C2 0.204105 0.1599448 C3 0.083971 0.0699759 0.07244 0.2930341 0.3512969 1 C4 0.083971 0.0552967 0.05470523 0.3238109 C5 0.254504 0.2284926 0.3140042 0.2586752 0.3248922 1 C6 0.069444 0.1183844 0.07667848 C7 0.02851 0.1693533 0.2157051 C8 0.156507 0.1985523 0.1007345 0.4482907 1 0.298622 C9 0.083179 Table 15 Initial supermatrix for distribution center Goal D1 D2 D3 D4 D5 D6 D7 D8 D9 Goal 1 1 1 1 1 1 1 1 1 D1 0.049747 0.1476923 D2 0.089082 0.3765877 0.449081 0.4117647 0.5837838 D3 0.110549 D4 0.042783 D5 0.059989 0.3780924 0.305924 0.4162162 D6 0.092585 D7 0.196225 D8 0.151759 0.5130919 0.3893707 0.5882353 1 D9 0.20728 0.1103204 0.0848445 0.2449951 312 Advances in Production Engineering & Management 14(3) 2019 A novel multiple criteria decision-making approach based on fuzzy DEMATEL, fuzzy ANP and fuzzy AHP for mapping ... Table 16 Normalized supermatrix for collection center Goal Cl C2 C3 C4 C5 C6 C7 C8 C9 Goal 0 0.5 0.500001973 1 0.5 0.5 0.4350433 0.5 0.5 1 CI 0.050884 0 0.082862627 0 0 0 0 0 0 0 C2 0.200914 0.0799724 0 0 0 0 0 0 0 0 C3 0.082658 0.0349879 0.036220143 0 0.14651705 0 0.1528293 0.5 0 0 C4 0.082658 0.0276484 0.027352723 0 0 0 0.1408717 0 0 0 C5 0.250525 0.1142463 0.157002719 0 0.1293376 0 0.1413422 0 0.5 0 C6 0.068359 0.0591922 0.038339391 0 0 0 0 0 0 0 C7 0.028064 0.0846767 0.107852975 0 0 0 0 0 0 0 C8 0.15406 0.0992762 0.050367449 0 0.22414535 0.5 0.1299135 0 0 0 C9 0.081878 0 0 0 0 0 0 0 0 0 Table 17 Normalized supermatrix for distribution center Goal D1 D2 D3 D4 D5 D6 D7 D8 D9 Goal 0 0.5 0.500000025 0.499999975 1 0.5 1 1 0.5 0.5 D1 0.049747 0 0.073846154 0 0 0 0 0 0 0 D2 0.089082 0.18829385 0 0.224540489 0 0.20588235 0 0 0.2918919 0 D3 0.110549 0 0 0 0 0 0 0 0 0 D4 0.042783 0 0 0 0 0 0 0 0 0 D5 0.059989 0 0.189046209 0.152961992 0 0 0 0 0.2081081 0 D6 0.092585 0 0 0 0 0 0 0 0 0 D7 0.196225 0 0 0 0 0 0 0 0 0 D8 0.151759 0.25654595 0.19468536 0 0 0.29411765 0 0 0 0.5 D9 0.20728 0.0551602 0.042422252 0.122497544 0 0 0 0 0 0 Table 18 Weights of criteria for collection and distribution center Collection center Distribution center Criteria Weights Criteria Weights CI 0.03817 D1 0.04272 C2 0.11587 D2 0.17471 C3 0.08293 D3 0.06626 C4 0.05699 D4 0.02564 C5 0.31248 D5 0.13016 C6 0.04509 D6 0.05549 C7 0.03149 D7 0.11761 C8 0.27100 D8 0.24528 C9 0.04598 D9 0.14212 Table 19 Location alternatives for the furniture industry Code Lot area Cost Zone Nearby commu- Accessibility Other information nity population Fl 302 square PHP 8,500,000 Commercial and is 88,704 The site alternative is accessible meters (total con- surrounded by some by at least 5 five minor roads. tractcost] residential areas These minor roads lead to major roads that surround the site alternative F2 156 square PHP 1,200,000 Heavily residential 112,755 The site alternative is accessible meters (total contract area with nearby by minor roads leading to subdi- cost] commercial establish- visions and other access roads ments F3 45 square PHP 14,400 Moderate commercial 73,032 The site is located along the road An advance rental of two meters (monthly rental area with residential of one of the city's landmark. The months and a security fee] zones street can be accessed through deposit equivalent to one two major roads month is needed F4 1,800 square PHP 174,000.00 Industrial zone with 99,598 The site is located along a minor An advance rental of meters (monthly rental residences road and can be accessed three months and a secu- fee] through two alternate roads rity deposit equivalent to three months is needed; the minimum lease term is one year, and a postdated check for the succeeding monthly rent is required_ Table 20 Local weights of each alternative with respect to a criterion for a collection center Alternative Cl C2 C3 C4 C5 C6 C7 C8 C9 Fl 0.402238 0.1585758 0.2669257 0.1594486 0.15457 0.1218488 0.161685 0.213737 0.224195 F2 0.212937 0.226141 0.1652397 0.2851311 0.368942 0.4092921 0.38924 0.261059 0.354206 F3 0.103026 0.1914138 0.2852078 0.1649846 0.268178 0.2737234 0.194983 0.180938 0.203212 F4 0.281799 0.4238694 0.2826269 0.3904356 0.20831 0.1951358 0.254092 0.344266 0.218387 Advances in Production Engineering & Management 14(3) 2019 313 Ocampo, Himang, Kumar, Brezocnik Table 21 Local weights of each alternative with respect to a criterion for a distribution center Alternative Dl D2 D3 D4 D5 D6 D7 D8 D9 Fl 0.2118785 0.1791128 0.1526729 0.1392201 0.1299249 0.1858912 0.3122054 0.1321161 0.2086329 F2 0.2858034 0.2574002 0.3518347 0.2862499 0.213301 0.2353964 0.1801965 0.246009 0.09176981 F3 0.3138274 0.3994122 0.257968 0.3120752 0.4214795 0.2956678 0.1849814 0.1877439 0.5506188 F4 0.1884908 0.1640748 0.2375244 0.2624548 0.2352947 0.2830446 0.3226166 0.434131 0.1489784 Table 22 Global weights of collection and distribution center for the furniture industry Alternatives Collection Ranking Alternatives Distribution Ranking Fl 0.192069 4 Fl 0.18003288 4 F2 0.297314 1 F2 0.22323544 3 F3 0.219826 3 F3 0.32561239 1 F4 0.290791 2 F4 0.27111927 2 0,35 ♦ Tabunok (F3) f, 0,3 - % ♦ Pagsuabungan (F4) = 0.25 c | ♦ Basak (F2) I °-2 "" S ♦ Mabolo (F1) on "3 £ 0,1 O ° 0,05 -■ 0 -I-1-1-1-1-1-1-1 0 0,05 0,1 0,15 0,2 0,25 0,3 0,35 Global Weights for Collection Centre Fig. 7 Furniture industry satisfaction map 6. Discussion and managerial implications Potential strategic locations of chief logistics functions such as collection and distribution centers are evaluated using the proposed fuzzy MCDM model with key results presented in the previous section. The succeeding sections provide thorough analyses of each aspect considered in selecting a collection and distribution center given the case study in Cebu, Philippines. 6.1 Collection center function For a collection center, government policies and regulations (C3), material availability (C5) and supply of product return (C8) established the most number of influenced criteria over the other while the quality of product return (C9) has no significance towards other criteria. Moreover, the capacity of the center (C1) and initial investment (C2) are observed as being the most influenced criteria. The results can be viewed and justified as follows: Government policies and regulations (C3). Government legislation has been identified by Sharma et al. (2016) [29] as an important factor in adopting remanufacturing in a developing country. A government's support regarding remanufacturing can either be a major driver for remanufacturing (Xiang and Ming, 2011 [78]), or a major roadblock (Sharma et al., 2016 [29]). The government has a great contributing factor as it can impose legislation that could engage people and organizations in environmentally sustainable activities such as remanufacturing. This shows the dependency of investment (C3), environmental collaboration with consumers (C4) and proper disposal (C6) toward government policies and regulations. Currently, the Philippines does not have specific laws regarding remanufacturing and reverse logistic. Poor implementation, budgetary issues, weak monitoring and implementation, and lack of political will at both local and national level hinder the full effect of the policies and regulations (Magtolis and Indab, 2008 [81]). To make up for the lack of specific laws on remanufacturing, it is ideal that the location of a facility has a proper local implementation of other environmental policies and regulations to increase the efficiency of the collection of EOL units for remanufacturing, and increase the awareness, cooperation, and collaboration of people. 314 Advances in Production Engineering & Management 14(3) 2019 A novel multiple criteria decision-making approach based on fuzzy DEMATEL, fuzzy ANP and fuzzy AHP for mapping ... Material availability (C5) and supply of product return (C8). It is difficult to predict the quantity of return of materials and products, therefore placing a collection center that has the minimum quantity uncertainty and maximum material availability is a major concern for decision-makers. This is a consideration highlighted by Serrano et al. (2013) [24]. The ability to collect sufficient EOL units is considered as a major economic driver of remanufacturing by Toffel (2004) [82] for its economic advantages. When material availability and supply of product return do not have a profitable opportunity for an organization, placing a collection center may not be feasible (Wojanowski et al., 2007, [17]). An interdependent relationship is observed between material availability (C5) and supply of product return (C8). Moreover, when material availability (C5) of EOL product is high, the supply of product return (C8) is also high; this implies a directly proportional relationship between both criteria. Material availability (C5) and supply of product return (C8) also affect the capacity of the center (C1), initial investment (C2), environmental collaboration with customers (C4), and proper disposal (C6). Quality of product return (C9). There are different quality levels of return for each EOL product (Xiaoyan, 2012 [9]). Proper inspection of units should be administered to carefully assess the products' viability for remanufacturing. Aras et al. (2008) [83] emphasized that organizations should carefully strategize since a high number of returns may have poor quality creating a challenge as most of the time, quality is unknown and uncertain. This is the reason that quality of product return (C9) does not exhibit any significant relationship with other criteria. The inclusion of this criterion in the framework supports the statement of Aras et al. (2008) [83] on the importance of considering quality in collecting EOL units. The capacity of the center (C1) and initial investment (C2). Other criteria must first be assessed in determining the probable capacity and initial investment of the center. The viability of an area in terms of other criteria is evaluated first to ensure that the area is operationally and strategically feasible before considering the required capacity and needed investment to establish a collection center. This shows that capacity is affected by material availability (C5), environmental collaboration with customers (C4), and supply of product return (C8). Additionally, both are interdependent towards one another. These criteria demonstrate a directly proportional relationship, that is, with a greater capacity of a facility, higher investment is needed (Rao et al., 2015 [84]). 6.2 Distribution center function For a distribution center, proximity to customers (D8) established the most number of influences over other criteria while the demand of the second market of the area (D2) is greatly affected by other criteria. Moreover, government policies and regulations (D4), distance to suppliers (D6), and transportation (D7) have no significant influence and are not substantially affected by other criteria. Proximity to customers (D8). Capturing the interest of customers creates a challenge to support remanufactured products primarily in the Philippine context where remanufactured products are usually associated with inferior quality and are considered as second hand or reused products. This highlights the need to locate a distribution center in the vicinity of customers. By having the facility strategically proximate to customers, awareness, convenience, and increase potential customers may be evident. Consequently, maximized sales and profitability will be attained. The demand for the secondary market of the area (D2). Attaining a high demand will entail consumers to recognize, accept and be aware of the importance of remanufacturing as the demand of the secondary market of the area (D2) is greatly influenced by environmental collaboration with customers (D5), and proximity to customers (D8). A strategically located facility that captures a high demand would signify a tremendous economic advantage for an organization. Mitra (2007) [85] has stated that the demand to support remanufactured products can be driven by the inherently lower prices of these products. This scenario applies in the context of the Philippines since the market in this country is price-sensitive. Government policies and regulations (D4). The consideration of government policies for a distribution center is in contrary to the disposition of a collection center. Little emphasis has been given by Philippine legislation regarding the selling and distribution of remanufactured prod- Advances in Production Engineering & Management 14(3) 2019 315 Ocampo, Himang, Kumar, Brezocnik ucts. As an effect, remanufactured products are not given specific consideration. The little emphasis on the distribution of remanufactured products can be identified as an effect of the lack of legislation and encouragement towards the collection of EOL units. This scenario is contrasting to the effect of policies in India determined by Govindan et al. (2016) [37, 38] where regulations towards EOL units restrict the flow of remanufactured products. With the lack of specific legislation towards collecting EOL units, it follows that the distribution aspect is not given importance as well. This is evident in the mathematical results of fuzzy DEMATEL where government policies and regulations (D4) is neither affected nor being affected by other criteria. Transportation (D7). Most of the time, manufacturers are unwilling to distribute the goods themselves; instead, prefer a third-party logistics provider to perform such operation (Govindan et al., 2012 [86]). For this reason, transportation (D7) forms no significant interrelationship with other criteria. Material availability (C5) the supply of product return (C8), and initial investment (C2). These criteria are deemed by decision-makers to be the most critical considerations in selecting a location for a collection center. A considerable gap is observed between the prioritization of these criteria and the remaining criteria. These criteria are perceived to be the significant drivers that motivate organizations; moreover, this prioritization reveals that economic and profitability concerns are significant considerations to set-up facilities for the collection of EOL units. On the other hand, proximity to customers (D8), the demand of the secondary market (D2) and land price (D9) are perceived to be the most important criteria in choosing a location for a distribution center. Similar to the selection of a location for a collection center, economic and profitability concerns are observed to be more prioritized. Basing on these trends, it can be inferred that economic sustainability is the primary driver for choosing a location alternative for both collection and distribution centers. 6.3 Evaluation of alternative locations Cebu is a leading exporter in the furniture industry. It accounts for 60 % of the country's total exports of the said sector, making Cebu the furniture capital of the country (PwC Cebu 2017 CEO Survey, 2017 [87]). With this favorable condition of the furniture industry in Cebu, the host company expressed its willingness to expand and improve its operations locally to increase its competitive advantage in the market. The case firm is open to the idea of setting up facilities for reverse logistics to enhance their operations. Four alternatives, F1, F2, F3, and F4 are critically assessed by decision-makers in terms of its viability as a collection and distribution center. The application of FAHP to the viability of locations for both collection and distribution centers in the furniture industry are discussed as follows. As a potential collection center, F2 exhibits the highest priority. It has been identified that the major contributing factors for this result are material availability (C5), proper disposal (C6), land price (C7), and quality of product returns (C9). From Table 22, it is observed that F2 is given more priority in terms of the proper disposal (C6) criterion. This alternative has a stricter implementation of policies and regulations towards proper waste management as the area is highly industrialized and is more pressured to comply with environmental regulations. The highly urbanized setting of F2 denotes that there is a high EOL unit availability in the area that can be collected. This could be explained in Table 22, as F2 ranked first regarding material availability (C5). F4 is believed to have the second-highest priority level and is considered as the most important alternative with regards to the initial investment (C2), environmental collaboration with customers (C4), and supply of product returns (C8). F4 ranks first in terms of supply of product returns (C8) since there is also a high priority in environmental collaboration with customers (C4). Notably, there is only a slight difference between the priority levels of F2 and F4. Although F4 ranks first in terms of initial investment (C2) and supply of product return (C8), which are second and third priority for the selection of location for collection center, decision-makers preferred F2 over F4 due to its high material availability (C5) which ranked first in the prioritization on the selection of location for collection center. Comparing the prioritized criteria for F4 and F2, F4 has more important criteria for a collection center. However, considering the remaining criteria, F2 demonstrates more priority. This is then succeeded by F3 and F1 with a high pri- 316 Advances in Production Engineering & Management 14(3) 2019 A novel multiple criteria decision-making approach based on fuzzy DEMATEL, fuzzy ANP and fuzzy AHP for mapping ... ority for government policies and regulations (C3) and capacity of the center (C1) respectively. In terms of the capacity of the center (C1), while F1 has a lesser lot area compared to F3, decision-makers prefer F1 as it is believed to have sufficient capacity for the operations of a collection center. With regards to a distribution center, F3 exhibits the highest priority. It has been determined that the key contributing factors in selecting a distribution center are distance of facility between competitor (D1), demand of the second market of the area (D2), government policies and regulations (D4), environmental collaboration with customers (D5), distance to suppliers (D6), and land price (D9). Land price (D9) is considered to be the most significant criterion and is also one of the top priority on the selection of a location for a distribution center, that impacts F3. The expense for the land acquisition of F3 is relatively lower than the other location alternatives making it more preferred by decision-makers. The current operations of the company are currently based in the northern area of Cebu, establishing a distribution center in F3 allows the company to venture in the southern area. Moreover, it allows them to test their profitability of penetrating in a new area. F3 also has the highest ranking of environmental collaboration with customers (D5) and demand of the second market of the area (D2). Exploiting the demand and acceptance of remanufactured products in F3 will be favorable for the company in terms of profitability. F4 is deemed to have the second-highest priority and is perceived to have the most straightforward transportation and the highest proximity to customers. Although proximity to customers (D8) is the top priority in selecting a location for a distribution center, F3, which has a relatively lower weight for proximity to customers (D8), was still favoured by decision-makers due to the other criteria that have a significant influence in setting up distribution center such as demand of the second market of the area (D2) and environmental collaboration with customers (D5). This is then followed by F2, which is significant in the investment criterion. The least preferred alternative for a distribution center is F1 without significance to any criteria. In summary, the Philippine furniture industry aims to be a global design innovator using sustainable materials by 2030. The industry intends to focus on factors such as product development where sustainable and environment-friendly materials are being used in manufacturing processes (DTI BOI, 2016 [88]). The government has been collaborating with private sectors to address the roadblock that hinders the growth of the manufacturers and improve policies for consistency and sustainability. The proposed framework of this study gives the government an insight as to the attractiveness of its policies among location alternatives. This allows an evaluation of a location's disposition towards policies for the collection of EOL units and distribution of remanufactured products. The government can impose additional policies or enforce stricter implementation of existing policies should they opt to make regions more supportive of remanu-facturing activities. The identification and prioritization of critical criteria allow the government to implement better and more precise legislation that focuses on improving a location's inclination on a specific criterion. A policy that can be considered by the government is providing incentives to organizations and or consumers through subsidies to organizations. An attractive incentive policy entices organizations to engage in remanufacturing activities and encourages consumers to return EOL units which increase the supply of product return (C8). Another action that can be performed by the government is initiating and enforcing laws that would regulate the remanufacturing sector to standardize their operations. This organizes the remanufacturing sector of the country which is comprised mostly of independent remanufacturers. Developing countries such as the Philippines lack heuristic research on reverse logistic studies specifically on the branch of remanufacturing. This paper can be considered as a pioneering study that discusses facility location planning supporting remanufacturing. Existing studies on facility location mostly utilizes a single objective criterion and fails to address other critical criteria that would affect the selection location for a collection and distribution center. Since the study is relatively new, it can start an interest among researchers who would like to further supplement the gap in the literature. The criteria determined for a collection and distribution center is applicable in general industry and can be used for future studies. Advances in Production Engineering & Management 14(3) 2019 317 Ocampo, Himang, Kumar, Brezocnik 7. Conclusion This study proposes a comprehensive approach to solve a facility location problem using fuzzy multiple criteria decision-making (MCDM) techniques. Since the facility location is one of the crucial problems that decision-makers encounter, it is necessary to assess the implications of establishing collection and distribution centers. The proposed approach provides a holistic decision that simultaneously considers multiple criteria that are critical for an EOL product collection center and a remanufactured product distribution center. It is determined in this paper that the government plays a vital role in the decision for the selection of facility locations. Since government policies and regulations (C3 and D4) have a significant interdependent relationship with the other criteria and are evaluated with the alternatives, the government is given an insight as to where locations are most and least preferred for policies. In the field of research in the Philippines, there is a lack of interest and studies on the subject of remanufacturing particularly in the location decision. This study is significant as it supplements this research gap. This study has identified nine essential criteria for both collection and distribution centers in the context of the Philippines. These criteria can be used by decision-makers as a reference in solving a location problem. The proposed framework provides decision-makers critical evaluation of location alternatives for facilities with consideration to the established criteria. The approach allows decision-makers to address major concerns regarding collecting EOL units and distributing remanufactured products. For example, in the collection function, quantity and quality uncertainties are significant factors that must be taken into consideration; while economic and market-oriented issues are major concerns for a distribution function. The methodology incorporates these various essential criteria that enable decision-makers to perform comprehensive judgment. Moreover, the framework enables decision-makers to assess the suitability of an alternative at strategic, tactical, and operational levels. The location alternatives are ranked based on their viability to perform reverse logistics functions. The decision-makers are given insights and can select a location to perform collection or distribution regardless of ranking as long as it fits the strategic plan of an organization. This allows decision-makers to evaluate specific location that can operate satisfactorily, that is, a location perceived to have sufficient performance to increase economic, social, and environmental opportunities. Remanufacturing in the Philippines is a mostly unappreciated industrial sector. This can be inferred from the country not having specific laws regarding remanufacturing and reverse logistics. The Philippines still has many issues that must be addressed, such as limited and ill-informed policies, cultural preferences, and assessment of actual benefits. The lack of legislation towards reverse logistic practices particularly in EOL unit collection affects the entire operation of remanufacturing. From the results of the surveys, government policies and regulations (C3 and D4) are seen as an important criterion as it is considered for both collection and distribution centers. Thus, the support of government is essential in improving the overall condition of re-manufacturing. The framework, in general, allows decision-makers to select a location that enables them to exploit the potential of a location as a collection and distribution center. This approach increases the economic and operational viability of reverse logistics operations. Decision-makers retain the control in picking a site based on its priority on being a collection or distribution center. The flexibility of the framework enables decision-makers to select a site as to their preference. For instance, decision-makers can choose to perform collection and distribution on a location for its practicality and applicability instead of its high viability depending on the strategic plan of the organization. Outside the reverse logistic literature, the methodological framework proposed in this work can be used to address general facility location problems where tradeoffs among stakeholders or entities are well pronounced and decision-makers find it imperative that such tradeoffs must be carefully observed. With a different application domain, a few changes in the criteria set, and the interrelationships of these criteria (i.e., other applications could set them as a priori) can be implemented. Future work could explore the following: (1) the hierarchical network problem could be better expounded by adding sub-criteria components in the criteria set, (2) the use of other MCDM methodologies, such as data envelopment analysis, best-worst meth- 318 Advances in Production Engineering & Management 14(3) 2019 A novel multiple criteria decision-making approach based on fuzzy DEMATEL, fuzzy ANP and fuzzy AHP for mapping ... od, multi-objective optimization, in the prioritization of the location alternatives could be carried out as fuzzy AHP imposes limits on the number of alternatives, and (3) the priority of the decision-makers as to the importance of collection and distribution functions is not reflected in the proposed methodological framework which could change the satisfaction trade-off map. Finally, the proposed approach could be set within the context of location-allocation problems. References [1] Mutha, A., Pokharel, S. (2009). Strategic network design for reverse logistics and remanufacturing using new and old product modules, Computers & Industrial Engineering, Vol. 56, No. 1, 334-346, doi: 10.1016/j.cie.2008. 06.006. [2] Sheu, J.-B. (2011). Bargaining framework for competitive green supply chains under governmental financial intervention, Transportation Research Part E: Logistics and Transportation Review, Vol. 47, No. 5, 573-592, doi: 10.1016/j.tre.2010.12.006. [3] Rashid, A., Asif, F.M.A., Krajnik, P., Nicolescu, C.N. (2013). Resource conservative manufacturing: An essential change in business and technology paradigm for sustainable manufacturing, Journal of Cleaner Production, Vol. 57, 166-177, doi: 10.1016/j.jclepro.2013.06.012. [4] Govindan, K., Diabat, A., Shankar, K.M. (2015). Analyzing the drivers of green manufacturing with fuzzy approach, Journal of Cleaner Production, Vol. 96, 182-193, doi: 10.1016/j.jclepro.2014.02.054. [5] Govindan, K., Soleimani, H., Kannan, D. (2015). Reverse logistics and closed-loop supply chain: A comprehensive review to explore the future, European Journal of Operational Research, Vol. 240, No. 3, 603-626, doi: 10.1016/ j.ejor.2014.07.012. [6] USITC (2012). Remanufactured goods: An overview of the U.S. and global industries, markets, and trade, United States International Trade Commission Publication, Investigation No. 332-525, USITC Publication 4356, Washington, USA. [7] Wei, S., Cheng, D., Sundin, E., Tang, O. (2015). Motives and barriers of the remanufacturing industry in China, Journal of Cleaner Production, Vol. 94, 340-351, doi: 10.1016/j.jclepro.2015.02.014. [8] Rathore, P., Kota, S., Chakrabarti, A. (2011). Sustainability through remanufacturing in India: A case study on mobile handsets, Journal of Cleaner Production, Vol. 19, No. 15, 1709-1722, doi: 10.1016/j.jclepro.2011.06.016. [9] Chen, J.-M., Chang, C.-I. (2012). The co-operative strategy of a closed-loop supply chain with remanufacturing, Transportation Research Part E: Logistics and Transportation Review, Vol. 48, No. 2, 387-400, doi: 10.1016/j.tre. 2011.10.001. [10] Xiaoyan, W. (2012). Research on design management based on green remanufacturing engineering, Systems Engineering Procedia, Vol. 4, 448-454, doi: 10.1016/j.sepro.2012.01.009. [11] Srivastava, S.K. (2006). Network design for reverse logistics, Omega, Vol. 36, No. 4, 535-548, doi: 10.1016/j. omega.2006.11.012. [12] Nnorom, I.C., Osibanjo, O. (2008). Overview of electronic waste (e-waste) management practices and legislations, and their poor applications in the developing countries, Resources, Conservation, and Recycling, Vol. 52, No. 6, 843-858, doi: 10.1016/j.resconrec.2008.01.004. [13] Zou, Z.-B., Wang, J.-J., Deng, G.-S., Chen, H. (2016). Third-party remanufacturing mode selection: Outsourcing or authorization?, Transportation Research Part E: Logistics and Transportation Review, Vol. 87, 1-19, doi: 10.1016/ j.tre.2015.12.008. [14] Malik, S., Kumari, A., Agrawal, S. (2015). Selection of locations of collection centers for reverse logistics using GTMA, Materials Today: Proceedings, Vol. 2, No. 4-5, 2538-2547, doi: 10.1016/j.matpr.2015.07.199. [15] Pop, P.C., Pinteaa, C.-M., Pop Sitar, C., Hajdu-Macelaru, M. (2015). An efficient reverse distribution system for solving sustainable supply chain network design problem, Journal of Applied Logic, Vol. 13, No. 2, Part A, 105113, doi: 10.1016/j.jal.2014.11.004. [16] Hong, I.-H., Yeh, J.-S. (2012). Modeling closed-loop supply chains in the electronics industry: A retailer collection application, Transportation Research Part E: Logistics and Transportation Review, Vol. 48, No. 4, 817-829, doi: 10.1016/j.tre.2012.01.006. [17] Atasu, A., Toktay, L.B., Van Wassenhove, L.N. (2013). How collection cost structure drives the manufacturer's reverse channel choice, Production and Operations Management, Vol. 22, No. 5, 1089-1102, doi: 10.1111/j.1937-5956.2012.01426.x. [18] Wojanowski, R., Verter, V., Boyaci, T. (2007). Retail-collection network design under deposit-refund, Computers & Operations Research, Vol. 34, No. 2, 324-345, doi: 10.1016/j.cor.2005.03.003. [19] Min, H., Ko, H.-J. (2008). The dynamic design of a reverse logistics network from the perspective of third-party logistics service providers, International Journal of Production Economics, Vol. 113, No. 1, 176-192, doi: 10.1016 /j.iipe.2007.01.017. [20] Paksoy, T., Pehlivan, N.Y., Kahraman, C. (2012). Organizational strategy development in distribution channel management using fuzzy AHP and hierarchical fuzzy TOPSIS, Expert Systems with Applications, Vol. 39, No. 3, 2822-2841, doi: 10.1016/j.eswa.2011.08.142. [21] Govindan, K., Chaudhuri, A. (2016). Interrelationships of risks faced by third party logistics service providers: A DEMATEL based approach, Transportation Research Part E: Logistics and Transportation Review, Vol. 90, 177195, doi: 10.1016/j.tre.2015.11.010. Advances in Production Engineering & Management 14(3) 2019 319 Ocampo, Himang, Kumar, Brezocnik [22] Chiou, C.Y., Chen, H.C, Yu, C.T., Yeh, C.Y. (2012). Consideration factors of reverse logistics implementation - A case study of Taiwan's electronics industry, Procedía - Social and Behavioral Sciences, Vol. 40, 375-381, doi: 10.1016/j.sbspro.2012.03.203. [23] Zadeh, L.A. (1965). Fuzzy sets, Information and Control, Vol. 8, No. 3, 338-353, doi: 10.1016/S0019-9958(65)90 241-X. [24] Serrano, C., Aggoune-Mtalaa, W., Sauer, N. (2013). Dynamic models for green logistic networks design, 1FAC Proceedings Volumes, Vol. 46, No. 9, 736-741, doi: 10.3182/20130619-3-RU-3018.00101. [25] Langevin, A., Mbaraga, P., Campbell, J.F. (1996). Continuous approximation models in freight distribution: An overview, Transportation Research Part B: Methodological, Vol. 30, No. 3, 163-188, doi: 10.1016/0191-2615(95) 00035-6. [26] Yang, X. (2013). A review of distribution related problems in logistics and supply chain research, International Journal of Supply Chain Management, Vol. 2, No. 4, 1-8. [27] Mukherjee, K., Mondal, S. (2009). Analysis of issues relating to remanufacturing technology - A case of an Indian company, Technology Analysis & Strategic Management, Vol. 21, No. 5, 639-652, doi: 10.1080/0953732090296 9174. [28] Choudhary, N., Singh, N.K. (2011). Remanufacturing in India: Approaches, potentials & technical challenges, International Journal of Industrial Engineering and Technology, Vol. 3, No. 3, 223-227. [29] Sharma, V., Garg, S.K., Sharma, P.B. (2016). Identification of major drivers and roadblocks for remanufacturing in India, Journal of Cleaner Production, Vol. 112, Part 3, 1882-1892, doi: 10.1016/j.jclepro.2014.11.082. [30] Yi, P., Huang, M., Guo, L., Shi, T. (2016). A retailer oriented closed-loop supply chain network design for end of life construction machinery remanufacturing, Journal of Cleaner Production, Vol. 124, 191-203, doi: 10.1016/ j.jclepro.2016.02.070. [31] Owen, S.H, Daskin, M.S. (1998). Strategic facility location: A review, European Journal of Operational Research, Vol. 111, No. 3, 423-447, doi: 10.1016/S0377-2217(98)00186-6. [32] Zhang, Y., Snyder, L.V., Ralphs, T.K., Xue, Z. (2016). The competitive facility location problem under disruption risks, Transportation Research Part E: Logistics and Transportation Review, Vol. 93, 453-473, doi: 10.1016/j.tre. 2016.07.002. [33] Kuo, M.-S. (2011). Optimal location selection for an international distribution center by using a new hybrid method, Expert Systems with Applications, Vol. 38, No. 6, 7208-7221, doi: 10.1016/j.eswa.2010.12.002. [34] Sarkis, J., Gonzalez-Torre, P., Adenso-Diaz, B. (2010). Stakeholder pressure and the adoption of environmental practices: The mediating effect of training, Journal of Operations Management, Vol. 28, No. 2, 163-176, doi: 10.1016/j.jom.2009.10.001. [35] Zhang, T., Chu, J., Wang, X., Liu, X., Cui, P. (2011). Development pattern and enhancing system of automotive components remanufacturing industry in China, Resources, Conservation, and Recycling, Vol. 55, No. 6, 613-622, doi: 10.1016/j.resconrec.2010.09.015. [36] Nunes, K.R.A., Mahler, C.F., Valle, R.A. (2009). Reverse logistics in the Brazilian construction industry, Journal of Environmental Management, Vol. 90, No. 12, 3717-3720, doi: 10.1016/j.jenvman.2008.05.026. [37] Govindan, K., Khodaverdi, R., Vafadarnikjoo, A. (2016). A grey DEMATEL approach to develop third-party logistics provider selection criteria, Industrial Management & Data Systems, Vol. 116, No. 4, 690-722, doi: 10.1108/ IMDS-05-2015-0180. [38] Govindan, K., Madan Shankar, K., Kannan, D. (2016). Application of fuzzy analytic network process for barrier evaluation in automotive parts remanufacturing towards cleaner production: A study in an Indian scenario, Journal of Cleaner Production, Vol. 114, 199-213, doi: 10.1016/j.jclepro.2015.06.092. [39] Ijomah, W.L., McMahon, C.A., Hammond, G.P., Newman, S.T. (2007). Development of design for remanufacturing guidelines to support sustainable manufacturing, Robotics and Computer-Integrated Manufacturing, Vol. 23, No. 6, 712-719, doi: 10.1016/j.rcim.2007.02.017. [40] Xu, L., Mathiayazhagan, K., Govindan, K., Noorol Haq, A., Ramachandran, N.V., Ashokkumar, A. (2013). Multiple comparative studies of green supply chain management: Pressures analysis, Resources, Conservation and Recycling, Vol. 78, 26-35, doi: 10.1016/j.resconrec.2013.05.005. [41] Saavedra, Y.M.B., Barquet, A.P.B., Rozenfeld, H., Forcellini, F.A., Ometto, A.R. (2013). Remanufacturing in Brazil: Case studies on the automotive sector, Journal of Cleaner Production, Vol. 53, 267-276, doi: 10.1016/j.jclepro. 2013.03.038. [42] Promentilla, M.A.B., Furuichi, T., Ishii, K., Tanikawa, N. (2008). A fuzzy analytic network process for multi-criteria evaluation of contaminated site remedial countermeasures, Journal of Environmental Management, Vol. 88, No. 3, 479-495, doi: 10.1016/j.jenvman.2007.03.013. [43] Tseng, M.-L. (2011). Using a hybrid MCDM model to evaluate firm environmental knowledge management in uncertainty, Applied Soft Computing, Vol. 11, No. 1, 1340-1352, doi: 10.1016/j.asoc.2010.04.006. [44] Tseng, M.L., Lin, Y.H., Chiu, A.S.F., Liao, J.C.H. (2008). Using FANP approach on selection of competitive priorities based on cleaner production implementation: A case study in PCB manufacturer, Taiwan, Clean Technologies and Environmental Policy, Vol. 10, No. 1, 17-29, doi: 10.1007/s10098-007-0109-4. [45] Fontela, E., Gabus, A. (1972). World problems an invitation to further thought within the framework of DEMATEL, Battelle Geneva Research Centre, Geneva. [46] Gabus, A., Fontela, E. (1973). Perceptions of the world problem atique: Communication procedure, communicating with those bearing collective responsibility, DEMATEL Report No.1, Battelle Geneva Research Centre, Geneva. 320 Advances in Production Engineering & Management 14(3) 2019 A novel multiple criteria decision-making approach based on fuzzy DEMATEL, fuzzy ANP and fuzzy AHP for mapping ... [47] Wang, R.-C., Chuu, S.-J. (2004). Group decision-making using a fuzzy linguistic approach for evaluating the flexibility in a manufacturing system, European Journal of Operational Research, Vol. 154, No. 3, 563-572, doi: 10.1016/S0377-2217(02)00729-4. [48] Tsai, W.-H., Chou, W.-C. (2009). Selecting management systems for sustainable development in SMEs: A novel hybrid model based on DEMATEL, ANP, and ZOGP, Expert Systems with Applications, Vol. 36, No. 2, Part 1, 14441458, doi: 10.1016/j.eswa.2007.11.058. [49] Büyüközkan, G., £iffi, G. (2012). A novel hybrid MCDM approach based on fuzzy DEMATEL, fuzzy ANP and fuzzy TOPSIS to evaluate green suppliers, Expert Systems with Applications, Vol. 39, No. 3, 3000-3011, doi: 10.1016/ j.eswa.2011.08.162. [50] Bongo, M.F., Ocampo, L.A. (2017). A hybrid fuzzy MCDM approach for mitigating airport congestion: A case in Ninoy Aquino international airport, Journal of Air Transport Management, Vol. 63, 1-16, doi: 10.1016/ j.jairtraman.2017.05.004. [51] Lin, R.-J. (2013). Using fuzzy DEMATEL to evaluate the green supply chain management practices, Journal of Cleaner Production, Vol. 40, 32-39, doi: 10.1016/i.iclepro.2011.06.010. [52] Baykasoglu, A., Kaplanoglu, V., Durmu§oglu, Z.D.U., $ahin, C. (2013). Integrating fuzzy DEMATEL and fuzzy hierarchical TOPSIS methods for truck selection, Expert Systems with Applications, Vol. 40, No. 3, 899-907, doi: 10.1016/i.eswa.2012.05.046. [53] Kumar, A., Mussada, E.K., Ashif, M., Tyagi, D., Srivastava, A.K. (2017). Fuzzy Delphi and hybrid AH-MATEL integration for monitoring of paint utilization, Advances in Production Engineering & Management, Vol. 12, No. 1, 4150, doi: 10.14743/apem2017.1.238. [54] Saaty, T.L. (1977). A scaling method for priorities in hierarchical structures, Journal of Mathematical Psychology, Vol. 15, No. 3, 234-281, doi: 10.1016/0022-2496(77)90033-5. [55] Saaty, T.L. (1996). Decision making with dependence and feedback: The analytic network process, RWS Publications, Pittsburgh, USA. [56] Shyur, H.-J. (2006). COTS evaluation using modified TOPSIS and ANP, Applied Mathematics and Computation, Vol. 177, No. 1, 251-259, doi: 10.1016/i.amc.2005.11.006. [57] Tavana, M., Khalili-Damghani, K., Abtahi, A.-R. (2013). A hybrid fuzzy group decision support framework for advanced-technology prioritization at NASA, Expert Systems with Applications, Vol. 40, No. 2, 480-491, doi: 10.1016/i.eswa.2012.07.040. [58] Chen, S.H., Wang, P.W., Chen, C.M., Lee, H.T. (2010). An analytic hierarchy process approach with linguistic variables for selection of an R&D strategic alliance partner, Computers & Industrial Engineering, Vol. 58, No. 2, 278287, doi: 10.1016/i.cie.2009.10.006. [59] Liao, C.-N. (2011). Fuzzy analytical hierarchy process and multi-segment goal programming applied to new product segmented under price strategy, Computers & Industrial Engineering, Vol. 61, No. 3, 831-841, doi: 10.1016/i.cie.2011.05.016. [60] Wu, C., Zhang, X.-Y., Yeh, I.-C., Chen, F.-Y., Bender, J., Wang, T.-N. (2013). Evaluating competitiveness using fuzzy analytic hierarchy process - A case study of Chinese airlines, Journal of Advanced Transportation, Vol. 47, No. 7, 619-634, doi: 10.1p02/atr.183. [61] Wang, Y., Jung, K.-A., Yeo, G.-T., Chou, C.-C. (2014). Selecting a cruise port of call location using the fuzzy-AHP method: A case study in East Asia, Tourism Management, Vol. 42, 262-270, doi: 10.1016/i.tourman.2013.11.005. [62] Dozic, S., Lutovac, T., Kalic, M. (2017). Fuzzy AHP approach to passenger aircraft type selection, Journal of Air Transport Management, Vol. 68, 165-175, doi: 10.1016/i.iairtraman.2017.08.003. [63] Banduka, N., Tadic, D., Macuzic, I., Crnjac, M. (2018). Extended process failure mode and effect analysis (PFMEA) for the automotive industry: The FSQC-PFMEA, Advances in Production Engineering & Management, Vol. 13, No. 2, 206-215, doi: 10.14743/apem2018.2.285. [64] Razmi, J., Rafiei, H., Hashemi, M. (2009). Designing a decision support system to evaluate and select suppliers using fuzzy analytic network process, Computers & Industrial Engineering, Vol. 57, No. 4, 1282-1290, doi: 10.1016/i.cie.2009.06.008. [65] Ayag, Z., Özdemir, R.G. (2009). A hybrid approach to concept selection through fuzzy analytic network process, Computers & Industrial Engineering, Vol. 56, No. 1, 368-379, doi: 10.1016/i.cie.2008.06.011. [66] Babaesmailli, M., Arbabshirani, B., Golmah, V. (2012). Integrating analytical network process and fuzzy logic to prioritize the strategies - A case study for tile manufacturing firm, Expert Systems with Applications, Vol. 39, No. 1, 925-935, doi: 10.1016/i.eswa.2011.07.090. [67] Uygun, Ö., Kafamak, H., Kahraman, Ü.A. (2014). An integrated DEMATEL and Fuzzy ANP techniques for evaluation and selection of outsourcing provider for a telecommunication company, Computers & Industrial Engineering, Vol. 86, 137-146, doi: 10.1016/i.cie.2014.09.014. [68] Lin, C.-L., Wu, W.-W. (2004). A fuzzy extension of the DEMATEL method for group decision making, European Journal of Operational Research, Vol. 156, 445-455. [69] Ocampo, L., Clark, E., Tanudtanud, K.V. (2015). Structural decisions of sustainable manufacturing strategy with fuzzy analytic network process (FANP), International Journal of Strategic Decision Sciences, Vol. 6, No. 2, 12-27, doi: 10.4018/iisds.2015040102. [70] Saaty, T.L. (2008). Decision making with the analytic hierarchy process, International Journal Services Sciences, Vol. 1, No. 1, 83-98, doi: 10.1504/IISSCI.2008.017590. [71] Mikhailov, L., Tsvetinov, P. (2004). Evaluation of services using a fuzzy analytic hierarchy process, Applied Soft Computing, Vol. 5, No. 1, 23-33, doi: 10.1016/i.asoc.2004.04.001. Advances in Production Engineering & Management 14(3) 2019 321 Ocampo, Himang, Kumar, Brezocnik [72] Wang, Y.-M., Chin, K.-S. (2011). Fuzzy analytic hierarchy process: A logarithmic fuzzy preference programming methodology, International Journal of Approximate Reasoning, Vol. 52, No. 4, 541-553, doi: 10.1016/i.iiar.2010. 12.004. [73] Krishnan, T.N., Poulose, S. (2016). Response rate in industrial surveys conducted in India: Trends and implications, IIMB Management Review, Vol. 28, No. 2, 88-97, doi: 10.1016/i.iimb.2016.05.001. [74] Kannan, G. (2009). Fuzzy approach for the selection of third party reverse logistics provider, Asia Pacific Journal of Marketing and Logistics, Vol. 21, No. 3, 397-416, doi: 10.1108/13555850910973865. [75] Govindan, K., Pokharel, S., Kumar, P.S. (2009). A hybrid approach using ISM and fuzzy TOPSIS for the selection of reverse logistics provider, Resources, Conservation and Recycling, Vol. 54, No. 1, 28-36, doi: 10.1016/i.resconrec. 2009.06.004. [76] Mittal, V.K., Sangwan, K.S. (2014). Prioritizing barriers to green manufacturing: Environmental, social and economic perspectives, Procedia CIRP, Vol. 17, 559-564, doi: 10.1016/i.procir.2014.01.075. [77] Ocampo, L.A., Promentilla, M.A.B. (2016). Development of a sustainable manufacturing strategy using analytic network process, International Journal of Business and Systems Research, Vol. 10, No. 2-4, 262-290, doi: 10.1504 /IIBSR.2016.075744. [78] Xiang, W., Ming, C. (2011). Implementing extended producer responsibility: Vehicle remanufacturing in China, Journal of Cleaner Production, Vol. 19, No. 6-7, 680-686, doi: 10.1016/i.iclepro.2010.11.016. [79] Andel, T., Aichlmayr, M. (2002). Turning returns into cash, Transportation Distribution, Vol. 43, No. 8, 29-39. [80] McAllister, J. (2015). Factors influencing solid-waste management in the developing world, All Graduate Plan B and other Reports, Utah State University, Logan, Utah, USA. [81] Magtolis, C.M., Indab, J.D. (2008). A policy study on environmental protection in the Philippines, Philippine Journal of Public Administration, Vol. 52, No. 2-4, 364-379. [82] Toffel, M.W. (2004). Strategic management of product recovery, California Management Review, Vol. 46, No. 2, 120-141, doi: 10.2307/41166214. [83] Aras, N., Aksen, D., Tanugur, A.G. (2008). Locating collection centres for incentive-dependent returns under a pick-up policy with capacitated vehicles, European Journal of Operational Research, Vol. 191, No. 3, 1223-1240, doi: 10.1016/i.eior.2007.08.002. [84] Rao, C., Goh, M., Zhao, Y., Zheng, J. (2015). Location selection of city logistics centers under sustainability, Transportation Research Part D: Transport and Environment, Vol. 36, 29-44, doi: 10.1016/i.trd.2015.02.008. [85] Mitra, S. (2007). Revenue management for remanufactured products, Omega, Vol. 35, No. 5, 553-562, doi: 10.1016/i.omega.2005.10.003. [86] Govindan, K., Palaniappan, M., Zhu, Q., Kannan, D. (2012). Analysis of third party reverse logistics provider using interpretive structural modeling, International Journal of Production Economics, Vol. 140, No. 1, 204-211, doi: 10.1016/i.iipe.2012.01.043. [87] PwC Cebu 2017 CEO Survey (2017), from https://www.pwc.com/ph/en/ceo-survey/2017/great-expectations-cebu-ceo-survey-2017.pdf. accessed November 13, 2018. [88] DTI BOI (2016). Securing the Future of Philippine industries: Furniture, from http://industry.gov.ph/industry/ furniture/, accessed November 13, 2018. 322 Advances in Production Engineering & Management 14(3) 2019 APEM jowatal Advances in Production Engineering & Management Volume 14 | Number 3 | September 2019 | pp 323-332 https://doi.Org/10.14743/apem2019.3.330 ISSN 1854-6250 Journal home: apem-journal.org Original scientific paper Effect of aluminium and chromium powder mixed dielectric fluid on electrical discharge machining effectiveness Modi, M.a*, Agarwal, G.b department of Mechanical Engineering, Acropolis Institute of Technology and Research, Indore, Madhya Pradesh, India department of Mechanical Engineering, Malaviya National Institute of Technology, Jaipur, Rajasthan, India A B S T R A C T A R T I C L E I N F O This article studied the impacts of using different powders on the productivity of electro discharge machining (EDM) of Nimonic 8OA alloy. The powders used for experiments are chromium (Cr) and aluminium (Al), though these powders are in contrasts in their thermo-physical characteristics. With the mixing of these powders in dielectric fluid, effect on surface roughness (SR), material removal rate (MRR), and mechanism of the machining process have been studied in this research work. On going through the results of experiments, it was observed that even volumetric proportion of powders, size of molecules, its density, electric resistance, and heat conductivity of additives were vital parameters that altogether influenced the productivity of powder mixed-electro discharge machining (PMEDM) process. With addition of proper ratio of powders in dielectric fluid, it enhanced the material removal rate, and consequently, reduced the surface roughness. Under a similar molecule volumetric proportion tests, the minutes suspended molecule size of powder prompted the largest material removal rate and consequently, the surface roughness increased. Conclusion is that adding chromium powder improves to the highest material removal rate, but poor surface finish while adding aluminium powder has the reverse effects. © 2019 CPE, University of Maribor. All rights reserved. Keywords: Powder mixed-electro discharge machining (PMEDM); Aluminium powder; Chromium powder; Dielectric fluid; Productivity; Material removal rate (MRR); Surface roughness; Nimonic 80A alloy *Corresponding author: manojmnitjaipur1@gmail.com (Modi, M.) Article history: Received 12 April 2019 Revised 13 September 2019 Accepted 15 September 2019 1. Introduction Electrical discharge machining (EDM) has been an essential procedure for the die and tool making enterprises for a long time. This machining procedure is finding growing utilization in industries due to the feasibility of machining geometrically intricate shapes irrespective of the hardness of materials that are very tough by utilizing the traditional machining procedures [1]. In EDM, managed distinct electrical sparks between the electrode and the work-piece would produce arcing, i.e. concentration of spark when an excessive amount of scattered tiny material remains in the inter-electrode gap in light of the inability to clear scattered tiny material. Arcing occurs when a succession of sparks strikes more than once on a similar spot. This, in the end, causes harm to the cathode of the apparatus and the work-piece if the controller of the machine does not stop it quickly. The EDM procedure in this manner ends up unsteady and ineffective. To enhance the effectiveness of ED machining, it is necessary to upgrade process steadiness. This presently remains a major tough task. Kiyak et al. [2] conducted experimental work on EDM with AISI P20 tool steel. They examined the impact of process factors on surface roughness (SR) in EDM and observed that littler values of ampere-current, pulse-on-duration and proportionately greater value of pulse-off-duration produced a descent surface 323 Modi, Agarwal finish (SF). Hasfalik et al. [3] did experimental work on EDM with Ti6Al4V and different tool materials. They investigated that the MRR, SR, electrode wear, and average white layer thickness (AWLT) ascend with increment in ampere-current and pulse-on-duration. They also examined the surface-integrity of work with EDM process factors. Fonda et al. [4] conducted experimental work on EDM with Ti6AL4V work-piece and reported the impact of properties of the material on EDM efficiency. Chow et al. [5] did experiments on micro-slit electrical discharge machining (MS EDM) with Ti6Al4V workpiece. They examined the impact of silicon carbide powder blended with dielectric fluid (DF) on MRR, SR, and gap size (GS). Chow et al. [6] conducted experiments on EDM with Ti6Al4V. They observed that the silicon-carbide and aluminium powder blended with kerosene is liable for increasing the gap-size among the electrodes. The developed GS escalates the debris removal rate (DRR), and material removal depth (MRD). They also examined the impact of adding the powders molecule in dielectric fluid on MRD and on the SR. Zhao et al. [7] did experimental work on EDM with steel as a workpiece and red copper as a tool material. They described the mechanism of EDM, and powder mixed electrical discharge machining (PMEDM) process and found that PMEDM process upgrades the machining efficiency and SF in contrast with EDM process by picking the suitable factors set. Pecas et al. [8] performed the experiments on PMEDM with and without silicon powder blended DF. They reported that the mixing of powder-molecule in DF alters few process parameters and makes the circumstances to accomplish the peak surface quality in vast regions. Pecas et al. [9] conducted experiments on PMEDM with AISI H13. They explained that silicon powder blended DF enhances the process polishing-effectiveness and reduces the white layer thickness (WLT), depth, and diameter of the cavity. They reported that exact-control of the flushing-rate and volumetric proportion of powder is the requirement for achieving the advancement in the shining ability of process. Kozak et al. [10] conducted experiments on EDM with tool steel. Experiments were conducted with kerosene and with powder blended demonized water. They narrated about the impact of process factors on MRR and SR with various volumetric proportions of powders in dielectric liquids. Kansal et al. [11] narrated that DF with powder in PMEDM process reduces the dielectric insulating-strength and enhances the gap among the tool and work. They also found that the addition of powder enhances the process-outcomes. Kansal et al. [12] did experimental work on PMEDM of AISID2 die-steel with silicon powder blended DF. They found that the chosen process factors have a remarkable impact on MR. Pecas et al. [13] carried out the experiments on EDM with AISI H13 work and copper as an electrode with and without silicon powder blended DF. The outcomes of the experiments accomplished confirm a straight connection among the electrode area and the surface quality measures and also a noteworthy execution enhancement when the powder blended DF is utilized. Modi et al. [14] described the detail of EDM and PMEDM process and developed the mathematical models of process responses through dimensional analysis (DA) method. Modi et al. [15] did the experiments on EDDSG with Ti6AL4V. They reported that Grey-TM methodology enhances machining effectiveness. Marashi et al. [16] reported the intensive review of the impact of powder addition on the mechanism of EDM process, the most influential powder parameters, future patterns in this technology, and a relative survey of powder materials is additionally exhibited in this article to facilitate a deeper insight into powder selection parameters for future investigations. They also portrayed that main factors which must be considered in PMEDM are powder size, type, and concentration. At last in this paper, PMEDM research patterns, gaps, findings, and industrial troubles are discussed extensively. Kalaman et al. [17] presented the study about the introduction and survey of the research work in PMEDM. The investigations concerning machining efficiency, surface integrity, and generation of functional surfaces are presented and discussed in the light of current research patterns. Attempts made to improve biocompatible surfaces with the utilization of the process additionally included clarifying the future patterns in PMEDM. Daneshmand et al. [18] conducted the experiments on EDM with CK45 steel to examine the impacts of current, voltage, and pulse frequency on SR. In this test work, kerosene is utilized as dielectric fluid and copper as an electrode. Design of experiments with non-linear regression model was utilized to estimate the process response. Raju et al. [19] carried out an extensive literature study to provide a total description on [¿-EDM process, its necessities, execution and 324 Advances in Production Engineering & Management 14(3) 2019 Effect of aluminium and chromium powder mixed dielectric fluid on electrical discharge machining effectiveness applications. The experiment frame-ups and its subsystems, trial studies and optimizing techniques, created micro feature and different applications are additionally described in this paper. In spite of the favorable outcomes, the EDM procedure with added substances is not yet utilized broadly in industry. One of the essential reasons is that numerous key factors of the PMEDM procedure, incorporating the mechanism of machining with different added substances, which are not surely known. The complicated nature of this procedure, particularly from the impacts of the thermo-physical characteristic of added powders, subsequently, justifies the extra investigation. The objective of the article is to make a precise investigation of the impacts of powder properties in DF on the effectiveness of ED Machining with the end goal to improve the applicability of the procedure in the industry. The release-transivity, particularly, determines the frequency of sparking that controls the whole MRR, though the powder-particles striking impact has an insignificant cutting impact providing primarily to the enhancement of the SF. 2. Materials, methods and experimental set-up This paper uses nomenclature given in Appendix A. Nimonic 80A nickel-chromium alloy is widely utilized in several industries due to its resistance against to corrosion and high strength applications because of excellent mechanical properties at eminent temperature. PMEDM is an efficient process for machining hard-to-cut material like Nimonic 80A. The composition of Nimonic 80A is 19.82 % Cr, 2.59 % Ti, 1.57 % Al, 2.63 % Fe, and balance % Ni. The PMEDM frame-up contains the servo arrangement of ZNC EDM machine (ZNC 50 x 30, die sinking type, EMT ltd., India) is utilized to keep up the predetermined separation among the tool and work, whose width rely on the gap voltage. Total tests were conducted on this frame-up with Nimonic 80A as a work-piece and bronze as an anode. For this reason, an independent fitting has been planned and created within the EDM tank; a different acrylic tank having the capacity of 36 liters of DF was settled on the EDM table with bolts. The different stirrer and pump arrangement are settled in the acrylic tank. The stirrer and pump arrangement is utilized for proper mixing and circulation of powder blended dielectric liquid in the inter-electrode gap. The schematic diagram of this frame-up is displayed in Fig. 1. Table 1 displays the EDM process factors. The thermal characteristics of aluminim, and chromium powder molecules are depicted in Table 2. Fig. 2(a) displays the mechanism of machining in PMEDM process and the occurrence of series discharge in PMEDM process is depicted in Fig. 2(b). Fig. 1: The schematic diagram of PMEDM frame-up Advances in Production Engineering & Management 14(3) 2019 325 Modi, Agarwal y / / "X X x—X workpiece [a] [b] Fig. 2 (a) Mechanism of machining in PMEDM processes, (b) Occurrence of series discharge \ \ \ Table 1: EDM process factors Electrode (+) Workpiece (-) Time of machining Dielectric fluid Flow rate of dielectric fluid Powder Powder sizes Volumetric proportion of powder molecules (cm3/l) Current I (A) Ton (^s) Duty cycle_ Bronze (0 = 10 mm) Nimonic 80A (0 = 20 mm) Half hour Kerosene 4 l/ min Aluminium, Chromium 0.080-0.090 nm and 15-20 ^m 0.30, 0.60, 1.20 2.0, 5.0 5, 30, 80 0.67 Table 2: Powders properties Powder Density p, (kg/cm3) Thermal conductivity À, W/(cm • K) Electrical resistivity p, (Ü • cm) Melting-point-temperature Tm, (K) Specific heat C, Cal/(kg • K) Chromium 7.16 x 10-3 0.67 2.60 2148 0.110 x 103 Aluminium 2.70 x 10-3 2.38 2.45 933 0.215 x 103 In these experiments, the levels of input parameters have been selected after the conduction of pilot experiments. The MRR, GS, and SR have been measured as the responses in this experimental work. We conducted the various experiments on PMEDM setup and made the graphs between the input variables and the output responses. From these graph, conclusions were drawn. 3. Results and discussion 3.1 Impacts of powder properties on gap size Fig. 3 displays the effect of mixing of Al and Cr powder particles in the DF on the GS. It was observed that the gap among the anode and cathode with Al powder is marginally more when contrasted with the Cr powder blended dielectric liquid was found attributable to its littler electrical-resistivity. This littlest GS would be accountable for extreme gas blast pressure with Cr powder. Furthermore, the Cr powder density is greater than the Al powder density, resulting in arcing rather sparking. Consequently, SR is higher with Cr when contrasted with Al powder. Al powder delivered the better SF trailed by Cr powder. 326 Advances in Production Engineering & Management 14(3) 2019 Effect of aluminium and chromium powder mixed dielectric fluid on electrical discharge machining effectiveness —— KerosenewithAlpowderl5-20|jm ..........KerosenewithAlpowder 80-90 |arn .....«.....KerosenewithCrpowder 15-20 |im .....*— KerosenewithCrpowder80-90|im 0 20 40 60 80 100 Pulse on Time (jjs) Fig. 3: The impact of mixing of Al, and Cr powders particle on the gap size (/ = 5 A, DC = 0.67, Volumetric proportion = 0.60 cm3/l) 3.2 Impacts of powder particle size on material removal rate Fig. 4 demonstrates the impact of two Al powder-molecule sizes at different volumetric proportions on MRR. It was seen that the appropriate mixing of powder-particles improves the machining productivity by additional settling of the electric-release. The enhancement in process durability occurred due to the fairly larger GS in inter-electrode gap. The outcomes additionally disclosed that the rise in the size of the powder particle resulted in a decrease in the enhancement of the MRR. This can be ascribed to both lesser electrical power density and a more probability of anomalous release. Moreover, when the size of the molecule was more than the spark-gap (about 06-55 |im), the efficiency of machining evidently decayed. The MRR outcomes for the minimal size of the molecule (80-90 nm) at different volumetric proportions of powder particles were even more than those with (15-20 |im) molecule size of powder particles. This is happened due to the existence of less spark gap among the electrodes. For the current at 5 A, and volumetric proportion of molecule more than 0.60 cm3/l, demonstrated a notable falling trend in MRR. This was because of the joined impacts of lower electrical-density, the lesser striking of powder particles, unevenly distribution of particles, and very low growth in the release-transitivity. ------- Current=5 A, 80-90 nm ------ - Current=5 A, 15-20|im * Current=2 A, 80-90 nm —X— Current=2 A, 15-20 iim 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Volumetric Proportion of Aluminum Powder (cm}/l) Fig. 4: The impact of particle size and volumetric proportion of aluminium powders on MRR (Ton = 30 us, DC = 0.67) Current = 5 A, VP = 0.3 0 cm3/l f — Current=5 A, VP = 0.6 cm3/l • Current=5A.VP=1.2cm3/l 0 10 20 30 40 50 60 70 80 Pulse-on-time (us) Fig. 5: The impact of Ton and volumetric proportion of 80-90 nm aluminium powders molecule on MRR (DC = 0.67) Advances in Production Engineering & Management 14(3) 2019 327 Modi, Agarwal 2.4 c" 2 .X--. +■•■ Current=5 A, VP = 0.3 0 cm3/l XX-+- --X--- Current=5 A,VP = 0.6cm3/l a 0.8 Current=5 A, VP = 1.2 cm3/l ^ 0.4 0 0 10 20 30 40 50 60 70 80 Pulse-on-time ((is) Fig. 6: The impact of Ton and volumetric proportion of 15-20 ^.m aluminium powders molecule on MRR (DC = 0.67] 3.3 Impacts of molecule concentration on material removal rate Fig. 4 additionally demonstrates that MRR was affected by the volumetric proportion of aluminium powders. The reasonable distinction in MRR is observed for the current at a higher level because of the generation of more spark energy as displayed in Figs. 4, 5, and 6. It was also observed from Fig. 4 that the highest MRR is obtained at 5 A current, 0.6 cm3/l of Al powders, Ton = 30 [is, and molecules sizes under 95 nm. This demonstrates that an ideal spark gap was acquired at this setting of the volumetric proportion of powder and Ton, which delivered the optimum factors setting of power density, striking of molecule and release-transitivity in the EDM procedure. Figs. 5 and 6 demonstrate that the impacts of the volumetric proportion of molecule of 80-90 nm and 15-20 [im aluminium powders particle fluctuated along with the Ton. At the lesser Ton of 5 [is, 0.6 cm3/l provide the good MRR, trailed by 1.2 cm3/l, and with 0.30 cm3/l being the inferior. Though, it may be found in Fig. 6 that at the 5.0 A current, when the Ton was raised to 30 [is, the impacts begun to alter with 15-20 [im aluminium powders particle. The volumetric concentration of 0.6 cm3/l still delivered the good MRR, however, 0.30 cm3/l was superior as compared to 1.2 cm3/l. This pattern proceeded with the rise in the Ton. When the time was raised to 80 [is, 0.30 cm3/l exceed 1.2 cm3/l in MRR outcomes for both 80-90 nm, and 15-20 [im aluminium powders particle. This was because of extra warming impacts due to the rising the Ton. More evacuated molten materials were consequently produced that in the long run decreased the performance of the volumetric proportion of 1.2 cm3/l in improving the procedure steadiness. The purpose behind the primary outcomes with 15-20 [im aluminium powders particle was believed to be because of the bigger molecule size and its impacts. It demonstrated that the bigger molecule size of added substances among the spark gap was increasingly delicate to evacuate the tiny work materials production. 3.4 Impacts of molecule characteristics on material removal rate Fig. 7 demonstrates the impacts of two volumetric proportions of Al, and Cr powders particle on the MRR. The test outcomes demonstrated that chromium generated the best MRR, and aluminium the least when the volumetric proportion of powder was < 1.2 cm3/l and I = 5.0 A. At the point, when the ampere-current was at 2.0 A, the distinction in MRR turned out to be little because of the less input of energy. Fig. 3 demonstrates that the spark gap for chromium is lesser than that for aluminium as clarified already. In principle, there was a somewhat higher power density and gas explosion pressure for chromium. Moreover, chromium powder density is double than the aluminium powder, ensuing in a powerful powder-particle collision. Additionally, the heat conductivity of aluminium powder is about 3 times bigger as compared to chromium powder, which showed that more energy is taken out by the aluminium powder blended dielectric liquid. Subsequently, Cr created the biggest MRR. 328 Advances in Production Engineering & Management 14(3) 2019 Effect of aluminium and chromium powder mixed dielectric fluid on electrical discharge machining effectiveness ------- Current=5 AwithCrpowder ---■-•- Current=5 Awith Al powder Current=2 Awith Al powder - —x..... Current=2 Awith Crpowder 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Volumetric Proportion of Aluminum Powder (cm3/') Fig. 7: The impact of ampere-current and volumetric proportion of 15-20 [j.m aluminium and chromium powders molecule on MRR (DC = 0.67, Ton = 30 [is] 3.5 Effects of particle size on surface roughness Fig. 8 demonstrates the impacts of two aluminium and chromium powder molecule sizes with different volumetric proportions on SR. The best sorts of powders for upgrading the smoothness of the surface are Al, Cr etc. Aluminium powder-particle creates a good surface for all the molecule proportions. It is predominantly because of the joined impacts of small electrical-resistivity, adequate thermal-conductivity, and less-density of aluminium. Less electrical-resistivity makes a high spark gap, good thermal-conductivity removes more heat, and less-density avoid the arcing among the electrodes. These impacts jointly lead to less density of electrical spark bringing the low gas blast, in this way just shallow holes are created on the machined surface. Al powder produced the best surface finish followed by the Cr powder. The best surface finish is obtained at the volumetric proportion of 1.2 cm3/l of aluminium powder with 15-20 [j.m particle size. —♦ - Current = 5 A, Cr powder, 80-90 jiuii ------ Current = 5 A, Cr powder, 15-20 |.im * Current =5 A, Al powder, 80-90 jun .....*.....Current = 5 A. Al powder, 15-20 jim 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 VolumetricProportion of Aluminum Powder (cm31) Fig. 8: The impact of particle-size and volumetric proportion of Al powders molecule on SR (DC = 0.67, Ton = 30 [is] 3.6 Effects of particle concentration on surface roughness As appeared in Fig. 9, it was seen that at the 5 A current, the aluminium powder volumetric proportion of molecule has built a major effect in SR due to the bigger MR. It was noticed that the biggest molecule volumetric proportion of 1.20 cm3/l provide the worst SF at Ton = 80 [is due to 5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 ---+--- Current = 5 A,VP = 0.30cm3/l -- - Current = 5 A,VP = 0.60cm3/l --*-- Current = 5A,VP = 1.2cm3/l 0 10 20 30 40 50 60 70 80 Pulse-on-time {(is) Fig. 9: The impact of Ton and volumetric proportion of 80-90 nm aluminium powders molecule on SR (DC = 0.67, I = 5 A] Advances in Production Engineering & Management 14(3) 2019 329 Modi, Agarwal its highest release transitivity. It is also noted from Fig. 9 that the volumetric proportion of powder molecule was not the main parameter controlling the SR in the EDM procedure. Alternately, Ton seemed by all account, to be the most vital parameter. 3.7 Impacts of molecule characteristics on surface roughness Fig. 10 demonstrates the impact of seven molecule volumetric proportion levels of aluminium, and chromium powder-particles on the SR. It was seen that aluminium powder produced the good SF than the chromium powder particle for both the 2 A as well as at 5 A current. This happens because the evacuation of more molten material from work surface in presence of Cr powder as compared to the presence of Al powder in DF. Subsequently, Cr generated the biggest SR. Moreover, chromium powder density is double than the aluminium powder, ensuing in a powerful powder-particle collision and result in arcing. Additionally, the heat conductivity of aluminium powder is about 3 times bigger as compared to chromium powder, which showed that more energy is taken out by the aluminium powder blended dielectric liquid. ♦ Current=5 AwithCr powder ■ --- Current=5 AwithAl powder Current=2AwithCr powder Current=2AwithAl powder 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Volumetric Proportion of Powder (cm3/!) Fig. 10: The impact of ampere-current and volumetric proportion of 15-20 |im aluminium, and chromium powders molecule on SR (DC = 0.67, and Ton = 30 |is) 4. Conclusions In view of the test outcomes, the following conclusions are drawn: • The molecule size, volumetric proportion of powders, its density, electrical-resistivity, and thermal-conductivity are the significant attributes of powders influencing performance of ED Machining process. • Due to the inclusion of powder particle in DF, the gap for spark was increased. • Spark gap varies with the mixing of powders of various sizes in the dielectric fluid. Spark gap increases with increasing size of powder molecules. The minimum rise in the gap is produced with a size of 80-90 nm powders and with 15-20 [j.m generating the highest gap. • Highest MRR was observed with powders of size 80-90 nm and with 15-20 [j.m generating the smallest. For the SF, the opposite pattern is noticed. • Higher spark gap was observed by adding aluminium powder particles followed by using chromium powder. • Best MRR was observed with adding chromium followed by aluminium powder. For SF, the opposite pattern was noticed. • The volumetric proportion on mixing the powder with the dielectric fluid in the ratio of 0.6 cm3/l gives best MRR outcomes. • With 2 A, and 5 A current, results are quite higher than the impact of molecule size, volumetric proportion of powder, and other factors. • The MRR and SR vary with different combination of molecule sizes, volumetric proportions of Al, and Cr powders in dielectric fluid in PMEDM process. • Experiments can be conducted by using different combination of materials, dielectric fluids, and also various powders to get different findings in the PMEDM process. 330 Advances in Production Engineering & Management 14(3) 2019 Effect of aluminium and chromium powder mixed dielectric fluid on electrical discharge machining effectiveness References [1] Abu Zeid, O.A. (1997). On the effect of electrodischarge machining parameters on the fatigue life of AISI D6 tool steel, Journal of Materials Processing Technology, Vol. 68, No. 1, 27-32, doi: 10.1016/S0924-0136(96)02523-X. [2] Kiyak, M., £akir, o. (2007). Examination of machining parameters on surface roughness in EDM of tool steel, Journal of Materials Processing Technology, Vol. 191, No. 1-3, 141-144, doi: 10.1016/i.imatprotec.2007.03.008. [3] Hasfalik, A., ^ayda?, u. (2007). Electrical discharge machining of titanium alloy (Ti-6Al-4V), Applied Surface Science, Vol. 253, No. 22, 9007-9016, doi: 10.1016/i.apsusc.2007.05.031. [4] Fonda, P., Wang, Z., Yamazaki, K., Akutsu, Y. (2008). A fundamental study on Ti-6Al-4V's thermal and electrical properties and their relation to EDM productivity, Journal of Materials Processing Technology, Vol. 202, No. 1-3, 583-589, doi: 10.1016/i.imatprotec.2007.09.060. [5] Chow, H.-M., Yang, L.-D., Lin, C.-T., Chen, Y.-F. (2008). The use of SiC powder in water as dielectric for micro-slit EDM machining, Journal of Materials Processing Technology, Vol. 195, No. 1-3, 160-170, doi: 10.1016/ i.imatprotec.2007.04.130. [6] Chow, H.-M., Yan, B.-H., Huang, F.-Y., Hung, J.-C. (2000). Study of added powder in kerosene for the micro-slit machining of titanium alloy using electro-discharge machining, Journal of Materials Processing Technology, Vol. 101, No. 1-3, 95-103, doi: 10.1016/S0924-0136(99)00458-6. [7] Zhao, W.S., Meng, Q.G., Wang, Z.L. (2002). The application of research on powder mixed EDM in rough machining, Journal of Materials Processing Technology, Vol. 129, No. 1-3, 30-33, doi: 10.1016/S0924-0136(02)00570-8. [8] Pe^as P., Henriques, E. (2003). Influence of silicon powder-mixed dielectric on conventional electrical discharge machining, International Journal of Machine Tools and Manufacture, Vol. 43, No. 14, 1465-1471, doi: 10.1016/ S0890-6955(03)00169-X. [9] Pefas, P., Henriques, E. (2008). Effect of the powder concentration and dielectric flow in the surface morphology in electrical discharge machining with powder-mixed dielectric (PMD-EDM), International Journal of Advanced Manufacturing Technology, Vol. 37, No. 11-12, 1120-1132, doi: 10.1007/s00170-007-1061-5. [10] Kozak, J., Rozenek, M., Dabrowski, L. (2003). Study of electrical discharge machining using powder-suspended working media, Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture, Vol. 217, No. 11, 1597-1602, doi: 10.1243/095440503771909971. [11] Kansal, H.K., Singh, S., Kumar, P. (2007). Technology and research developments in powder mixed electric discharge machining (PMEDM), Journal of Materials Processing Technology, Vol. 184, No. 1-3, 32-41, doi: 10.1016/ i.imatprotec.2006.10.046. [12] Kansal, H.K., Singh, S., Kumar, P. (2007). Effect of silicon powder mixed EDM on machining rate of AISI D2 die steel, Journal of Manufacturing Processes, Vol. 9, No. 1, 13-22. doi: 10.1016/S1526-6125(07)70104-4. [13] Pe^as, P., Henriques, E. (2008). Electrical discharge machining using simple and powder-mixed dielectric: The effect of the electrode area in the surface roughness and topography, Journal of Materials Processing Technology, Vol. 200, No. 1-3, 250-258, doi: 10.1016/i.imatprotec.2007.09.051. [14] Modi. M., Jha, S. (2009). Modeling and analysis of powder mixed electric discharge machining, International Journal of Mechanical Engineering, Vol. 2, 219-223. [15] Modi, M., Agarwal, G. (2013). Optimization of electro-discharge diamond surface grinding process parameters with multiple performance characteristics of Ti-6Al-4V using grey-Taguchi approach, Advanced Materials Research, Vol. 622-623, 14-18, doi: 10.4028/www.scientific.net/AMR.622-623.14. [16] Marashi, H., Jafarlou, D.M., Sarhan, A.A.D., Hamdi, M. (2016). State of the art in powder mixed dielectric for EDM applications, Precision Engineering, Vol. 46, 11-33, doi: 10.1016/i.precisioneng.2016.05.010. [17] Kalaman, S., Ya§ar, H., Ekmekci, N., Opoz, T.T., Ekmekci, B. (2018). Powder mixed electrical discharge machining and biocompatibility: A state of the art review, In: Proceedings of the 18th International Conference on Machine Design and Production, Eski§ehir, Turkey, 803-830. [18] Daneshmand, S., Neyestanak, A.A.L., Monfared, V. (2016). Modelling and investigating the effect of input parameters on surface roughness in electrical discharge machining of CK45, Tehnicki Vjesnik - Technical Gazette, Vol. 23, No. 3, 725-730, doi: 10.17559/TV-20141024224809. [19] Raiu, L., Hiremath, S.S. (2016). A state-of-the-art review on micro electro-discharge machining, Procedia Technology, Vol. 25, 1281-1288, doi: 10.1016/i.protcy.2016.08.222. Advances in Production Engineering & Management 14(3) 2019 331 Modi, Agarwal Appendix A The following nomenclature is used in the paper: AWLT Average white layer thickness DC Duty cycle DF Dielectric fluid DRR Debris removal rate EDDSG Electro discharge diamond surface grinding EDM Electrical discharge machining GRA Grey relational analysis GS Gap size I Current (A) MR Machining rate MRD Material removal depth MRR Material removal rate (mm3/min) MS EDM Micro-slit EDM PMEDM Powder mixed electrical discharge machining Ra Average roughness of surface (|m) SF Surface finish SR Surface roughness TM Taguchi method Ton Pulse on-time (|s) WEDM Wire electrical discharge machining WLT White layer thickness 332 Advances in Production Engineering & Management 14(3) 2019 Advances in Production Engineering & Management Volume 14 | Number 3 | September 2019 | pp 333-342 https://doi.Org/10.14743/apem2019.3.331 ISSN 1854-6250 Journal home: apem-journal.org Original scientific paper Multi-objective scheduling of cloud manufacturing resources through the integration of Cat swarm optimization and Firefly algorithm Du, Y.a*, Wang, J.L.b, Lei, L.c aAnyang Institute of Technology, Anyang, P.R. China bGuangdong Pharmaceutical University, Guangzhou, P.R. China cZhengzhou University, Zhenzhou, P.R. China A B S T R A C T A R T I C L E I N F O This paper attempts to minimize the makespan and cost and balance the load rate of the process scheduling of cloud manufacturing resources. For this purpose, a multiobjective scheduling model was established to achieve the minimal makespan, minimal cost and balanced load rate. Next, the cat swarm optimization (CSO) and the firefly algorithm (FA) were combined into a hybrid multi-objective scheduling algorithm. Finally, the hybrid algorithm was verified through CloudSim simulation. The simulation results show that the algorithm output the optimal scheduling plan in a short time. This research not only provides an effective way to find the global optimal solution, within the shortest possible time, to the process scheduling problem of cloud manufacturing resources with multiple objectives, but also promotes the application of swarm intelligence algorithms in job-shop scheduling problems. © 2019 CPE, University of Maribor. All rights reserved. Keywords: Cloud manufacturing; Multi-objective scheduling; Cat swarm optimization (CSO); Firefly algorithm (FA) *Corresponding author: 421319725@qq.com (Du, Y.) Article history: Received 2 August 2019 Revised 8 September 2019 Accepted 10 September 2019 1. Introduction Cloud manufacturing is a web-based, service-oriented intelligent manufacturing paradigm. This concept was proposed by Li Bohu, Academician of Chinese Academy of Engineering, in 2010, with the aim to fully integrate social manufacturing resources, improve resource utilization, lower manufacturing cost and respond faster to market demand. Cloud manufacturing combines such techniques as cloud computing, the Internet of Things (IoT), high-performance computing and intelligent science. In this way, manufacturing resources and capacity can be managed and scheduled in a centralized, uniform and intelligent manner, and the resources can be allocated and scheduled more effectively. In a word, cloud manufacturing pursues "the centralized management of scattered resources and the distribution services of centralized resources". The cloud manufacturing service platform provides users with the required manufacturing resources and their full lifecycle services in real time. The process scheduling of cloud manufacturing resources depends on makespan, cost and load rate. Therefore, the cloud manufacturing task was decomposed into several processes, the basic units of scheduling. On this basis, a multi-objective scheduling model was established for the minimal makespan, minimal cost and balanced load rate. Meanwhile, a hybrid multi-objective scheduling algorithm was developed, coupling the cat swarm optimization (CSO) and the firefly algorithm (FA). CloudSim simulation shows that our algorithm outperformed the con- 333 Du, Wang, Lei trastive algorithms in makespan and search ability. Finally, an example was cited to prove that our algorithm can converge to the optimal scheduling plan in a short time, providing a desirable solution to multi-objective scheduling of cloud manufacturing resources. 2. Literature review In cloud computing, the scheduling problem involves cost, makespan as well as load balance. The cloud computing system decomposes the computing task into several subtasks, in the light of the massive data on the task. Next, each subtask was further split into processes. The more refined the division, the better the monitoring and control of the processing state. Thus, the process division ensures the professionality of service-oriented cloud manufacturing. Zhou et al. [1] sets up a mathematical model for multi-objective disassembly line balancing problem that minimizes the times of tool replacement, and puts forward a discrete tracking mode based on sequence exchange; Next, the CSO was integrated with the simulated annealing (SA) algorithm to enhance the global search ability; Finally, the proposed model and the hybrid algorithm were applied to design the disassembly line of a printer, creating various balanced solutions for decision makers [2-3]. Zuo et al. [4] establishes a cloud computing resource scheduling model based on improved chaotic FA. Firstly, the cloud computing resource scheduling model was built to improve the completion time, efficiency and safety of the task; Then, the chaotic algorithm was introduced to the FA, and the individuals were perturbed to speed up the convergence and avoid the local minimum trap. In addition, the Lagrangian relaxation function was adopted to improve the established model. CloudSim simulation shows that the improved chaotic FA can balance the resource allocation, shorten the completion time and enhance the system capacity. Sun et al. [5] mentions that cloud manufacturing resources raise a high requirement on scheduling and its granularity, due to their dispersity, diversity and load rate imbalance. In the light of this, the cloud manufacturing task was divided into processes, which were taken as the basic units of scheduling. Then, a multi-objective cloud manufacturing process scheduling model was constructed to minimize the makespan, minimize the cost and balance the load rate. After that, a hybrid multi-objective scheduling algorithm was designed based on the particle swarm optimization (PSO) and the genetic algorithm (GA) [6-7]. The bi-level coded chromosomes of the GA were taken as the particles of the PSO. Bi-level coding refers to the crossover and mutation of chromosomes through two layers, which speed up the convergence to the global optimum. The first layer is the sequence of processes and the second layer is the number of the resources corresponding to the processes. Finally, the hybrid algorithm was applied to schedule the cloud manufacturing of an elevator. The results show that the algorithm can output the optimal scheduling plan in a short time, and thus effectively solve multi-objective scheduling of cloud manufacturing. Wu et al. [8] designs an algorithm that establishes the mapping relationship between position vector of each particle and the allocated service through integer coding. The crossover and mutation operations of the GA were introduced to update the particle positions by standard PSO. The particle positions were updated by four methods in turns to ensure swarm diversity. Example analysis shows that the algorithm enjoys a high effectiveness and execution efficiency [9-12]. 3. Used methods 3.1 Cat swarm optimization The CSO models the behavior of cats into two modes: seeking mode and tracing mode. The former mainly performs local search and the latter looks for the global optimum. Under the seeking mode, the individuals were perturbed multiple times, such that each can approach the local optimum. Under the tracing mode, each cat traces the target at a certain speed, and updates its position into the better between its current position and the optimal position of the swarm [13]. Here, the tracing mode of the CSO is optimized. The current position of each cat was updated continuously according to the global optimal position [14]. Thus, the current solution can gradu- 334 Advances in Production Engineering & Management 14(3) 2019 Multi-objective scheduling of cloud manufacturing resources through the integration of Cat swarm optimization and ... ally approximate and reach the optimal solution. The speed and position update strategies can be expressed as: Vi(t + 1) = Vi(t) + ctxrand x (x* -%i(t)) (1) zi(t + 1) = zi(t) + vi(t + 1) (2) where x* is the current best-known position of the swarm; c1 is the acceleration coefficient; rand is a random number in [0, 1]; vt (t) and vt (t + 1) are the cat speed at the t-th and (t + 1)-th iterations, respectively; xt(t) and xt(t + 1) are the cat position at the t-th and (t + 1)-th iterations, respectively. Under the seeking mode, the speed parameter was removed because this mode only performs local search. Then, the position update strategy can be expressed as: Xi(t + 1) = Xi(t) + Timax (13) where Timax is the deadline on the delivery time of F^. The total cost of all subtasks must be smaller than the project budget. /2 >Qmax (14) where Cimax is the maximum budget of F^. 336 Advances in Production Engineering & Management 14(3) 2019 Multi-objective scheduling of cloud manufacturing resources through the integration of Cat swarm optimization and ... 5. Design of an integrated Cat swarm optimization and Firefly algorithm 5.1 Coding design Each of the i subtasks corresponds to Ij processes and k manufacturing resources. Thus, the feasible solutions can be coded with the positions of i xj xk fireflies. Then, the subtasks can be arranged into a sequence, in which each subtask must appear for j times. Let the subtask sequence be M pu),i, M(3,1),1, M(2,1),1, Mi(i,2),2, M(4,i),i, Mp,2p, M(i,3),3, Mp,3p, M(4,2),3, where Mjt, means subtask i should be processed in process j on manufacturing resource k. This sequence can be interpreted as follows: firstly, the first processes of the first, second and third subtasks should be processed in turns on the first manufacturing resource; next, the second process of the first sub-task should be processed on the second manufacturing resource. Finally, the second process of the fourth subtask should be processed on the third manufacturing resource [26]. 5.2 Swarm initialization According to the code design, the proposed CSO-FA algorithm code consists of two parts. The first part explains the processing sequence of subtask processes on the manufacturing resources and the second part specifies the number of the resources corresponding to the processes. The code length is (j x j) + k. Next, a swarm of N fireflies was initialized, and each firefly was assigned a random initial position in the given feasible region. 5.3 Fitness function The computer algorithm for cloud manufacturing should be designed in view of the specific demand of users. If the products are urgently needed (e.g., receiving a rush order with delay penalty), there is no need to think too much about cost and balance of load rate. The only goal to be pursued is to minimize the makespan. If the order is not urgent but with a small amount, it is only necessary to consider the cost. If the order faces limited manufacturing resources, the balance of load rate should be the top priority. Of course, overall consideration should be given to cost, makespan and load rate for most orders. In this paper, two types of fitness functions are discussed: the total fitness function and the sub-fitness function [27-28]. The total fitness function provides an evaluation criterion for scheduling plans. Each plan was represented as a firefly, whose brightness reflects the quality of the current position. The relationship between firefly position and brightness was established. Then, the brightness was taken as the value of the fitness function to evaluate each scheduling plan. The greater the fitness, the brighter the firefly, the more suitable the position, and the better the scheduling plan. The value of the total fitness function for multiple objectives can be derived from the random weighting of the single objectives: f, n^ ^^-^min , minC-Cmin ( minL-Lmin fitness (k) = ^---+ m2---+ ----(15) -'min ^min ^min where = —^— is the weight (r is a non-negative random number); Tmin, Cmin and Lmin are 2j=i ri the minimal makespan, minimal cost and balanced load rate of the three single-objective functions, respectively. The sub-fitness functions can be expressed as: T ■ 1 min = fitness^k) = mmT[(i,j),k],k = 1,2,.. ..,K (16) c = fitness2(k) = mmC[(i,j),k],k = 1,2,. ..,K (17) ^min = fitness3(k) = mmL[(i,j),k],k = 1,2,. ..,K (18) 5.4 Algorithm flow Based on the CSO-FA algorithm, a cloud computing task can be scheduled in the following steps: Step 1: Initialize the firefly swarm, determine the positions of M fireflies as per the problem scale, and set the algorithm parameters. Advances in Production Engineering & Management 14(3) 2019 337 Du, Wang, Lei Step 2: Judge if the maximum number of iterations has been reached. If so, end the iterative process; otherwise, proceed with the following steps. Step 3: Decode the coded fireflies, find the mapping relationship between manufacturing resources and the subtasks, and compute the minimum makespan, minimum cost and optimal load rate. Step 4: Randomly assign weight to each single-objective function, and compute the value of the total fitness function, i.e. the firefly brightness. Step 5: Let all fireflies move towards the nearby brighter individuals. Divide the search space into small regions. Apply the seeking mode of the CSO in the relatively bright regions, and the tracing mode of the CSO in the relatively dark regions. Update the position of each firefly to speed up the search. Step 6: Control the swarm diversity with equations. Step 7: Decode the global optimal firefly and output the result as the optimal scheduling plan. 6. Results and discussion The proposed CSO-FA algorithm was verified through simulation on CloudSim 4.0. CloudSim is a cloud computing simulation software released on April 8, 2009 by the Cloud Computing and Distributed Systems Laboratory, The University of Melbourne, Australia, in association with the Gridbus Project. It is a function library developed based on SimJava, a discrete event simulation package. The software can operate on both Windows and Linux. CloudSim inherits from GridSim the support to the R&D of cloud computing, and provides two distinctive new features: (1) the ability to model and simulate large cloud computing infrastructure; (2) the provision of a self-sufficient platform supporting data centers, service agents, as well as scheduling and allocation strategies. CloudSim also boasts many unique functions. For example, a virtualization engine is designed to help data centers provide multi-layered virtualization services both independently and collaboratively, and the processors assigned to visualization services can switch flexibly in time and space. The CSO-FA, the FA and the improved FA (IFA) were separately applied to simulate the scheduling of 5 manufacturing resources and 1,000 cloud computing tasks. The convergence curves of the three algorithms are shown in Fig. 1. Obviously, the FA saw a gradual slowdown of convergence speed, and converged prematurely in the global search, owing to its weak search ability. Compared with the FA, the IFA converged to the optimal solution rapidly. However, the fastest convergence and optimal solution were achieved by the CSO-FA. This is because the CSO-FA introduces the seeking mode and tracing mode to different regions, which speeds up the search for the global optimum and prevents the local optimal trap (the Y-axis is the convergence of form Fig. 1 to Fig. 4). 3 2.5 . ... z S<-... i.5 v;--... 1 \ 0.5 0 0 50 100 200 itera.numm 250 300 400 FA 2.7 2.5 2.3 1.2 1.3 1.25 1.25 * - * • IFA 2.45 1.75 1.3 0.8 0.65 0.5 0.4 _ CSO-FA 2.3 1.5 0.8 0.6 0.45 0.3 0.2 Fig. 1 The convergence curves of different algorithms 338 Advances in Production Engineering & Management 14(3) 2019 Multi-objective scheduling of cloud manufacturing resources through the integration of Cat swarm optimization and ... Table 1 Requested subtasks and their processes Subtask Subtask Number of Subtasks correspond Timax (s) Cimax (Yuan) number name processes to processes Fi floor 6 101,102,103,104,105,106 336 158 F2 Mounting plate 6 201,202,203,204,205,206 288 208 F3 Shock absorber 5 301,302,303,304,305 240 350 F4 Limit board 4 401,402,403,404 200 340 F5 Licating piece 5 501,502,503,504,505 192 258 F6 Card 6 601,602,603,604,605,606 160 267 FA ¡FA CSO-FA 80 100 120 140 160 180 200 itera.numm 94.9 94.5 93.7 93.2 93 92.9 92.8 92.7 92.5 92.4 94.7 94.4 93.5 93 92.7 92.5 92.3 91.9 91.6 91.5 94.6 94.2 93 92.5 92.4 92 91.7 91.3 91 90.7 Fig. 2 The makespans of different algorithms with 50 tasks 400 1.25 0.4 0.2 —X-- FA I FA CSO-FA 20 40 60 80 100 120 140 160 180 Itera.numm 95 94.9 94.5 93.7 93.2 93 92.9 92.8 92.7 92.5 92.4 95 94.7 94.4 93.5 93 92.7 92.5 92.3 91.9 91.6 91.5 95 94.6 94.2 93 92.5 92.4 92 91.7 91.3 91 90.7 Fig. 3 The makespans of different algorithms with 100 tasks Table 2 Available manufacturing resources Fa Fi2 Fi3 Fi4 Fi5 Fi6 M tijk Cijk M tijk Cijk M tijk Cijk M tijk Cijk M tijk Cijk M tijk Cijk F1 5 3 11 6 10 25 4 9 14 [2,9] [5,4] [9,11] [3,7] [3,3] [7,6] 10 4 8 F2 4 6 15 [2,9] [8,6] [12,9] 8 4 8 [6,7] [2,6] [5,8] 53 4 [1,10] [3,3] [7,7] F3 3 4 8 [6,8] [5,7] [10,9] 7 7 11 [2,1] [5,5] [15,16] [4,10] [9,11] [21,21] F4 5 7 16 2 3 7 [4,7] [4,6] [8,11] 10 3 8 F5 [4,5] [6,4] [9,14] 5 10 18 [9,10] [7,9] [17,19] 6 8 16 25 13 F6 [2,6] [3,7] [8,13] 4 10 20 [6,9] [8,7] [13,15] 7 9 15 84 7 [3,9] [9,4] [15,13] Table 3 Load rate of each manufacturing resource Ms M1 M2 M3 M4 M5 Me M7 M8 M9 M10 Ls 0.91 0.89 0.83 0.94 0.70 0.79 0.93 0.71 0.88 0.95 Advances in Production Engineering & Management 14(3) 2019 339 Du, Wang, Lei 242 217 0 20 40 60 " " " " " ~" 140 160 180 200 itera.numm FA ; 244 235 234 233.9233.8233.7233.7233.8233.8233.7233.7 FA ; 244 230 229.5229.4 229 228.5228.4228.3228.2 228 228 CSO-FA 244 227 225 224 223 220 220 220 219 219 219 Fig. 4 Iterative process of each algorithm 6.1 Comparison of makespans Without changing the number of manufacturing resources, the three algorithms were further compared through the simulation on 50 and 100 cloud computing tasks. The computing power of each manufacturing resource was randomly determined in the interval of [400, 200], and the subtask length was controlled within [400, 1,000]. The swarm size and number of iterations of the three algorithms were kept the same. The other parameters were configured in reference to related literature. Each algorithm was run 20 times and the mean value was taken as the final result. The simulation results are plotted as Figs. 2 and 3. As shown in Figs. 3 and 4, the proposed CSO-FA had a much shorter makespan than the FA and the IFA. This advantage is attributable to the strategy to control swarm diversity, which protects the search ability of the swarm and reduces the chance of falling into the local optimum trap. Hence, our algorithm is more suitable for cloud manufacturing scheduling than the contras-tive algorithms. 6.2 Example analysis The proposed algorithm was applied to schedule the manufacturing subtasks of an elevator enterprise on the cloud platform. There are a total of six manufacturing subtasks. The processes, delivery time Timax and budget Cimax of each subtask are listed in Table 1. For each subtask, the manufacturing resources suitable for its processed were searched for in the resource pool on the cloud platform. The pool provides 10 manufacturing resources for these subtasks. Table 2 describes the manufacturing resources M available for the processes of each subtask, and the makespan ttjk and cost Cijk of each process on different manufacturing resources. Table 3 specifies the load rate of each manufacturing resource. The iterative process of each algorithm is recorded in Fig. 4. 7. Conclusion Taking processes as the basic scheduling unit, this paper establishes a multi-objective optimization model for cloud manufacturing resource scheduling, in the light of the main influencing factors of cloud manufacturing scheduling. Besides, the CSO-FA for subtask scheduling in cloud computing was designed to rationalize the resource allocation in the cloud environment. Specifically, the CSO was introduced to the FA to accelerate the search process, without sacrificing the search ability of the swarm. The CSO-FA was applied to the established model, minimizing the time to converge to the global optimum. Compared with the FA and IFA, the proposed algorithm converged to the optimal solution in a short time. Finally, the algorithm was proved suitable to solve the multi-objective scheduling of cloud manufacturing resources, through the simulation of manufacturing order processing in an elevator enterprise. 340 Advances in Production Engineering & Management 14(3) 2019 Multi-objective scheduling of cloud manufacturing resources through the integration of Cat swarm optimization and ... Acknowledgement Fund support: Anyang science and technology research project [2018] No. 66. Key scientific research projects of institutions of higher learning in Henan province in 2020 Item no.20B460002. References [1] Zou, B.S., Zhang, Z.Q., Cai, N., Zhu, L.X. (2018). Cat swarm simulated annealing algorithm for disassembly line balancing problem under tool constraints, Computer Integrated Manufacturing Systems, Vol. 24, No. 9, 82-94, doi: 10.13196/j.cims.2018.09.009. [2] Huang, X., Zhang, T., Deng, Z., Li, Z. (2018). Design of moving target detection and tracking system based on cortex-A7 and openCV, Traitement du Signal, Vol. 35, No. 1, 61-73, doi: 10.3166/TS.35.61-73. [3] Zhang, D. (2017). High-speed train control system big data analysis based on fuzzy RDF model and uncertain reasoning, International Journal of Computers, Communications & Control, Vol. 12, No. 4, 577-591, doi: 10.15837/ ijccc.2017.4.2914. [4] Zuo, Z.-L., Guo, X., Li, W. (2018). An improved swarm optimization algorithm, Microelectronics and Computer, Vol. 35, No. 2, 61-66, doi: 10.19304/j.cnki.issn1000-7180.2018.02.013. [5] Sun, W., Wu, H., Lv, W., Gao, Y. (2017). Research on multi-objective process scheduling of cloud manufacturing resources, Journal of Nanjing University of Aeronautics and Astronautics, Vol. 49, No. 6, 773-778, doi: 10.16356/ j.1005-2615.2017.06.003. [6] Li, B.-H., Zhang, L., Wang, S.-L., Tao, F., Cao, J.-W., Jiang, X.-D., Song, X., Chai, X.-D., (2010). Cloud manufacturing: A new service-oriented networked manufacturing model, Computer Integrated Manufacturing System, Vol. 16, No. 1, 1-7, doi: 10.13196/j.cims.2010.01.3.libh.004. [7] Wang, S.-L., Song, W.-Y., Ling, K, Qiang, L.-I., Liang, G., Chen, G.-S. (2012). Manufacturing resource allocation based on cloud manufacturing, Computer Integrated Manufacturing System, Vol. 18, No. 7, 1396-1405. [8] Wu, Q., Ma, Y.-M., Li, L., Qiao, F. (2015). Data driven dynamic scheduling method for semiconductor production line, Control Theory and Application, Vol. 32, No. 9, 1233-1239. [9] Liu, W., Liu, B., Sun, D. (2013). Multi-task oriented service composition in cloud manufacturing, Computer Integrated Manufacturing System, Vol. 19, No. 1, 199-209. [10] Yin, C., Zhang, Y., Zhong, T. (2012). Optimization model of cloud manufacturing services resource combination for new product development, Computer Integrated Manufacturing System, Vol. 18, No. 7, 1368-1378. [11] Kshirsagar, P., Jiang, D., Zhang, Z. (2016). Implementation and evaluation of online system identification of electromechanical systems using adaptive filters, IEEE Transactions on Industry Applications, Vol. 52, No. 3, 2306-2314, doi: 10.1109/TIA.2016.2515994. [12] Puttonen, J., Lobov, A., Cavia Soto, M.A., Martinez Lastra, J.L. (2015). Planning-based semantic web service composition in factory automation, Advanced Engineering Informatics, Vol. 29, No. 4, 1041-1054, doi: 10.1016/ j.aei.2015.08.002. [13] Cao, Y., Wu, Z., Zhao, Q., Yan, H., Yang, J. (2015). Study on resource configuration on cloud manufacturing, Mathematical Problems in Engineering, Vol. 2015, Article ID 635602, 13 pages, doi: 10.1155/2015/635602. [14] Cao, Y., Wang, S., Kang, L., Gao, Y. (2016). A TQCS-based service selection and scheduling strategy in cloud manufacturing, The International Journal of Advanced Manufacturing Technology, Vol. 82, No. 1-4, 235-251, doi: 10.1007/s00170-015-7350-5. [15] Xiong, Y., Wang, J., Wu, M., She, J. (2015). Virtual resource scheduling method of cloud manufacturing oriented to multi-objective optimization, Computer Integrated Manufacturing System, Vol. 21, No. 11, 3079-3087, doi: 10.13196/j.cims.2015.11.029. [16] Ge, P., Yang, X., Xiao, X., Qiu, Y.-Q., Ren, P.-Y., Liu, Z.-S., Zheng, W.-M. (2012). Task allocation in cloud manufacturing based on multi-resolution clustering, Computer Integrated Manufacturing System, Vol. 18, No. 7, 1461-1468. [17] Wu, S.-Y, Zhang, P., Li, F. (2015). Multi-task scheduling based on particle swarm optimization in cloud manufacturing systems, Journal of South China University of Technology (Natural Science Edition), Vol. 43, No. 1, 105-110, doi: 10.3969/j.issn.1000-565X.2015.01.017. [18] Gao, Y., Tao, L., Ma, M. (2018). Multilevel-thresholding image segmentation based on cat swarm optimization, Chinese Stereology and Image Analysis, Vol. 23, No. 2, 125-132, doi: 10.13505/j.1007-1482.2018.23.02.001. [19] Du, Y., Shi, F., Chen, Q., Wang, Y., Zhao, J., Li, Q. (2018). An improved particle swarm scheduling algorithm based on batch changing production time, European Journal of Electrical Engineering, Vol. 20, No. 4, 439-453, doi: 10.3166/ejee.20.439-453. [20] Du, Y., Chen, Q.X. (2012). The model and simulation of SMT production, Applied Mechanics and Materials, Vol. 217-219, 1493-1496, doi: 10.4028/www.scientific.net/AMM.217-219.1493. [21] Pajoutan, M., Golmohammadi, A., Seifbarghy, M. (2014). CMS scheduling problem considering material handling and routing flexibility, The International Journal of Advanced Manufacturing Technology, Vol. 72, No. 5-8, 881-893, doi: 10.1007/s00170-014-5696-8. [22] Capps, O., Seo, S.-C., Nichols, J.P. (1997). On the estimation of advertising effects for branded products: An application to spaghetti sauces, Journal of Agricultural and Applied Economics, Vol. 29, No. 2, 291-302, doi: 10.1017/S1074Q70800007793. Advances in Production Engineering & Management 14(3) 2019 341 Du, Wang, Lei [23] Zacharia, P.T., Nearchou, A.C. (2016). A population-based algorithm for the bi-objective assembly line worker assignment and balancing problem, Engineering Applications of Artificial Intelligence, Vol. 49, 1-9, doi: 10.1016/ j.engappai.2015.11.007. [24] Miyake, D.I. (2006). The shift from belt conveyor line to work-cell based assembly systems to cope with increasing demand variation in Japanese industries, International Journal of Automotive Technology and Management, Vol. 6, No. 4, 419-439, doi: 10.1504/IIATM.2006.012234. [25] Sakazume, Y. (2004). Is Japanese cell manufacturing a new system? A comparative study between Japanese cell manufacturing and cellular manufacturing, Journal of Japan Industrial Management Association, Vol. 55, No. 6, 341-349. [26] Cui, J. (2018). The application of improved bacteria foraging algorithm to the optimization of aviation equipment maintenance scheduling, Tehnicki Vjesnik - Technical Gazette, Vol. 25, No. 4, 1103-1109, doi: 10.17559/TV-20171214025731. [27] Liu, Y., Wang, L., Wang, X.V., Xu, X., Zhang, L. (2019). Scheduling in cloud manufacturing: State-of-the-art and research challenges, International Journal of Production Research, Vol. 57, No. 15-16, 4854-4879, doi: 10.1080/ 00207543.2018.1449978. [28] Dalfard, V.M., Ranjbar, V. (2012). Multi-projects scheduling with resource constraints & priority rules by the use of simulated annealing algorithm, Tehnicki Vjesnik - Technical Gazette, Vol. 19, No. 3, 493-499. 342 Advances in Production Engineering & Management 14(3) 2019 APEM journal Advances in Production Engineering & Management Volume 14 | Number 3 | September 2019 | pp 343-354 https://doi.Org/10.14743/apem2019.3.332 ISSN 1854-6250 Journal home: apem-journal.org Original scientific paper Evaluation of the sustainability of the micro-electrical discharge milling process Pellegrini, G.a, Ravasio, C.a* aUniversity of Bergamo, Department of Management, Information and Production Engineering, Dalmine (BG), Italy A B S T R A C T A R T I C L E I N F O The sustainability evaluation of an industrial process is an actual issue: a process should not only grant part quality and high production rates at the lowest cost, but it should minimize its impact on the environment as well. Micro-EDM (Electrical Discharge Machining) is widely used in micro machining for its small force and high precision and environmental aspects related this technology are taken into account In this paper, an evaluation of the micro-ED milling process concerning the sustainability manufacturing was made. For this purpose, a method to assess the sustainability process was developed, taking into account the energetic consumption, the environmental impact, the dielectric consumption, the wear of the electrode and the machining performance. This method was applied for the execution of micro-pockets using two workpiece materials, two types of electrode and five types of dielectric, both liquid and gaseous. This analysis permits the identification of the critical aspects of the micro-ED milling process form the point of view of the sustain-ability. The comparison between the different solutions in terms of electrode material and dielectric underlines interesting considerations about the usage of non-traditional dielectrics. As regards electrode material, the environmental impact process when brass electrode is adopted is lower than tungsten carbide electrode. As concerns dielectric, water reveals to be the most sustainable dielectric; vegetable oil and oxygen, proved to be valid substitutes to traditional dielectrics under several viewpoints, including sustainability. © 2019 CPE, University of Maribor. All rights reserved. Keywords: Electrical discharge machining (EDM); Micro-electrical discharge machining (micro-EDM); Micro-electrical discharge milling (micro-ED milling); Sustainability; Sustainability index; Dielectric fluid *Corresponding author: chiara.ravasio@unibg.it (Ravasio, C.) Article history: Received 29 May 2019 Revised 13 September 2019 Accepted 16 September 2019 1. Introduction When dealing with process selection, sustainability issues are achieving increasing importance. A process should not only grant part quality and high production rates at the lowest cost, but it should minimize its impact on the environment as well. Whereas quality, productivity and cost can be evaluated using accepted techniques, sustainability evaluation is still matter of research. The simplest way of assessing sustainability is through indicators [1], although a more complex method of life cycle assessment has been proposed by international standards [2, 3]. In both cases, data collected from either production practice or experimental tests are supplied to a model which makes them dimensionally homogeneous and then aggregates them, generally by means of weighting factors. A scalar value (index) is then evaluated and used to rank different process conditions. To define a process index, knowledge about the subject is necessary to identify the factors [4]: in general, energy requirements (for both the main process and auxiliary actions), material usage (for both work and consumables), process fluids management (such as coolants, lubricants and dielectrics), waste production, health and safety issues should be taken into account. Among many other processes, sustainable production through electrical discharge 343 Pellegrini, Ravasio machining (EDM) has been studied [5-7]. Albeit a relatively small number of parts are currently produced via EDM, when compared to metal cutting operations, EDM is one of the most energy intensive processes [8]. Similar considerations can be extended to micro-EDM, by which small but increasing volumes are processed and the high specific power consumption is higher [9, 10]. In EDM, material removal is effected by a sequence of electrical discharges between work-piece and a conductive tool (electrode), removing small portions of material from both sides, i.e. from both work and tool. To improve the process, such discharges take place in a dielectric medium, whose purpose is reducing the spark size, localizing the energy supply, contributing to end the spark and to remove the vaporized material [11]. Two main EDM techniques became popular, namely die-sink EDM (often simply quoted as EDM) and Wire EDM (WEDM). EDM processes show many interesting properties, since they are able to produce complex shapes and they are not affected by the mechanical strength of the workpiece. On the other hand, some negative factors limit the field of application of such technique. First, the material removal rate (MRR) is critically low: for this reason, a good deal of research has been focused on improving the MRR [12, 13]. Moreover, tool wear rate (TWR) is always non negligible if compared with MRR, affecting material consumption and geometrical accuracy. Then, the process output depends on several parameters; process optimization is required to achieve good results but it is often nontrivial. As said, when assessing process performance for EDM operations, MRR and tool wear are mainly taken into account. Other important facts are geometrical accuracy and surface integrity. Recently, energy consumption and health impact are considered as well [14]. EDM can be used to machine small features (having characteristic size of about 1 mm or less), in this case the term micro-EDM is often used. Micro-EDM is based on the same physic principle as all EDM operations, yet each discharge conveys less energy and the discharge frequency is higher [15]. In this way, improved accuracy (small gaps are mandatory for small parts) is reached and the energy effectiveness is generally smaller, since the fraction of energy used for vaporizing the dielectric is relatively higher [16, 17]. Several types of operations are used in micro-EDM: micro-die sink and wire EDM are similar to their macro-scale counterpart, whereas some specific processes include micro-ED drilling, micro-ED milling, micro-ED grinding and micro-wire ED grinding (micro-WEDG). In micro-ED milling a rotating tool (of simple geometry) is moved with respect to the work as in traditional end milling, such as pocket milling or contouring applications. Compared with other micro-EDM processes, it achieves better flushing and tool stability [18]. In micro-ED milling, machining times are often quite long, but machining performance may be improved by either optimizing process parameters [19-21] or exploiting other techniques, as it has been reported in [22]. For both EDM and micro-EDM, process optimization may be carried out in several ways, by considering machining parameters or by selecting suitable materials for either electrode or dielectric fluid. Electrical parameter optimization involves the selection of current, voltage, spark time and spark interval. For this purpose, regression techniques are often used [23]. In general, commercial EDM devices are provided with machining programmes, including an optimal combination of parameters as a function of materials (for both work and tool) and of the surface requirements (either roughing or finishing); thus, fine tuning of electrical parameters is seldom feasible by the end user, who can only select among a set of machining programmes. In case of micro-EDM, a RC generator is very often employed: due to the physics of the power circuit, discharge current and duration cannot be independently chosen, so further reducing the degrees of freedom for optimization [8]. Further chances for optimization are provided by the selection of process materials, for both the dielectric and the electrode. Dielectric selection is dealt with in many studies [24]. At present, kerosene and deionized water (especially for WEDM) are mainly used [14, 25]. Desirable properties for a dielectric fluid are low specific gravity, high flash point and oxygen content, low viscosity and toxicity, high breakdown voltage and biodegradability [26]. Other solutions include organic oils, aerosols and gases [27, 28]. The use of a vegetable oil proved to achieve significant improvements with respect to kerosene in terms of MRR, surface finish and integrity, besides being more desirable for sustainability issues [6]. Gas assisted EDM (sometimes referred to as dry EDM) have been studied by several Authors. Both air [29] and 344 Advances in Production Engineering & Management 14(3) 2019 Evaluation of the sustainability of the micro-electrical discharge milling process oxygen [30] were studied, interesting findings have been reported in the last two decades. MRR, TWR, surface quality and integrity may be improved by using dry EDM [11]. When evaluating dielectric performances, health problems have been often considered. Dielectric fluids may release vapours and fumes; they may require a treatment (filtering, deioniz-ing) and they become exhausted over time. It has been pointed out that a quantitative assessment of health impact requires more data about pollutant concentration in fumes that is heavily dependent on dielectric type [31, 32]. Dielectric circulation and filtering generally use a significant fraction of the overall power consumption [10]. In this way, both sustainability and process cost are affected. It is worth noting that energy consumption in EDM is reported to be the main source of both cost and sustain-ability issues. A relevant part of energy consumption is independent from machining parameters [33], so the main impact on both cost and sustainability is due to machining time; thus, MRR consideration are of primary relevance. The sustainability evaluation of the micro-EDM process is a complex task because several aspects have to be taken into account Sustainability assessment through indexes may be the simplest way, yet it requires knowledge about process data. In the present paper a sustainability index is presented for micro-ED milling and used to compare different choices of dielectric fluids, both liquid and gas. An experimental campaign was carried out to supply data to the model. Five dielectrics, two work materials (titanium and stainless steel) and two electrode materials (brass and tungsten carbide) were taken into account for the study. 2. Materials and method 2.1. Development of the Sustainability Index The idea of sustainable manufacturing is not yet fully defined due to the presence of several interpretations of the sustainability concept [34]. In fact, several domains can be considered so, as a function of these, the expression sustainable manufacturing may assume different meanings. In this paper, the concept of sustainability is strictly connected to the manufacturing operations and the aspects related to stakeholders, technologies, services, supply chain are not included. In this view, the issues that are taken into account to evaluate the overall environmental impact are: the energy consumption, the materials requirement, the waste management, the process safety and the staff health (Fig. 1). A Sustainability Index (SI), expressed in euro, was developed representing the environmental impact in terms of quantity of the consumed resources and pollution effects created by the EDM machining. In this way, when the index assumes high values, the process is poorly sustainable, while for low values the process has a lower impact and therefore is more sustainable. Energy consumption Waste management Fig. 1 Sustainable manufacturing issues Advances in Production Engineering & Management 14(3) 2019 345 Pellegrini, Ravasio The elements taken into account to formulate SI are: Energetic consumption: It represents the electrical energy consumption of the micro-EDM machine. The energetic sustainability (Senergy) is calculated as follows: Senergy Etot ^el (1) where Etot represents the absorbed energy of the machine expressed in [kWh] and cei the cost per unit for the electricity in [€/kWh]. Electrode wear: The sustainability related to the electrode wear (Sw) can be evaluated as: Sw = W • ce (2] where W is the volume of the consumed electrode in [mm3] and ce is the unit cost per volume of the tool expressed in [€/mm3]. Dielectric: To estimate the impact of the dielectric, the purchase cost of both the dielectric and the filters and the costs for its disposal were taken into account. The purchase cost of the dielectric (Q) expressed in [€], charged to an EDM operation having a time duration of te (erosion time], was evaluated as: r _ Cgd '^d'^e ,„, Ld — 7 (3] Ld where Cad is the price per litre of the dielectric in [€/l], Vd is the volume of the dielectric tank [l] and Ld is the lifetime of the dielectric in [h]. It must be noted that the erosion time, expressed in [h], is time interval from the start to the end of machining cycle and therefore includes time period in which the machine does not erode (for example the time used to control the level of wear of the tool). Strictly speaking, only the active time should be taken into account but considering that the duration of the milling operation is much longer than the passive time, assuming that this latter is negligible leads to an acceptable approximation. This information is in agreement with [10] found for die-sinking EDM of macro components. The purchase cost of the filter (C^) in [€] for the dielectric unit can be estimated as: (4) Lf where caf is the cost of the filter in [€] and Lf is its lifetime expressed in [h]. It is worth noting that filter management involves the use of significant amount of energy; in this model, however, such value has already been included in overall energy consumption and therefore is not accounted here. The dismantling cost of the dielectric in [€] is defined as: c9 'V,j 'tp Cs = S j (5) Ld where cs is the unitary dielectric dismantling cost for liter in [€/l]. The dielectric sustainability takes into account the three over mentioned elements: Xcad + CS) 'Vd _ Co£ w Noted that the total effect of dielectric on sustainability is proportional to machining time and therefore it is directly dependent on MRR. Sdielectric — Q + + Q — ( £ ^ J^e [6] Process performance: This factor is related to the sustainability impact of scrap production that can be evaluated by multiplying the scrap rate a (i.e. the probability of producing a nonconforming part) by the average cost of either disposing or repairing the part. If a part cannot be repaired, its disposal cost is evaluated by taking into account the cost of the electrode wear, of the machine time and of the raw workpiece. When a part can be repaired, the associated cost is lower than the disposal cost. On average, the cost for producing a nonconforming part (to be either scrapped or repaired) is a fraction y of the total disposal cost. Scrap rate should be esti- 346 Advances in Production Engineering & Management 14(3) 2019 Evaluation of the sustainability of the micro-electrical discharge milling process mated through a statistical analysis; for the reported experiments, however, a simplified technique, based on deviations of the machined slot depth from its nominal value, was preferred. A suitable smoothing function was used to link deviations to scrap probability. On this basis, the performance sustainability can be estimated as follows: where cm is the hourly cost of the micro-EDM machine, expressed in [€/h], and k describes the value of the workpiece, for sake of simplicity, its value was set to zero. Environmental impact: It represents an evaluation of the environmental impact of the used dielectric. Several aspects were taken into account such as the fire hazard, the generation of fumes and vapours, the possible skin irritation of the operators, the generation of toxic fumes, the dust formation, the possible re-use of the dielectric and, finally, the dielectric and filters dismantling. For each of them, a qualitative evaluation from 0 to 3 was made based on literature data [7, 35] (increasing the value, the environmental impact is more severe). The penalty coefficient ki was defined as the ratio between the sum of these evaluations and the maximum achievable points. The environmental sustainability (Se) was evaluated as follows: The developed Sustainability Index is affected by the process performance, in terms of machining time and electrode wear, experimentally evaluating. 2.2. Experimental cases A Sarix SX-200 machine was used to realize micro-milling tests. Micro-pockets on two types of workpiece materials and using two different electrodes were machined. Fig. 2 shows the dimensions of the pocket having depth 0.1 mm. The milling strategy was layer-by-layer and the depth of each layer was 0.003 mm, adopting a roughing energy. When dealing with milling, the machine builder allows to select among some built in sets of process parameters (machining strategy, i.e. roughing, finishing etc.). For the present case, the machining strategy labelled roughing was selected. It can be noted that within each strategy, the electrical parameters depend on the workpiece material, the electrode characteristics and the type of dielectric (only kerosene and water are included). The machining parameters affect strongly the machining performance in term of machining time, electrode wear and geometrical characteristics. In its turn, the sustainability index depends strongly on machining time and on electrode wear (see the equations of the sub-indexes). The workpiece materials were stainless steel (AISI 316L) and titanium (Ti6Al4V); as regards the tool, tubular electrodes made of two different materials, tungsten carbide (WC) and brass, having external diameter of 0.3 mm and internal diameter of 0.12 mm, were used. As regards the dielectric, five types of dielectric were used, both liquid and gaseous: Kerosene (HEDMA 111), demineralized water, vegetable oil (soya bean), air at 10 bar and oxygen at 9.5 bar. Their properties are reported in Table 1. The gaseous dielectrics were injected in the machining zone thorough the tubular electrode (Fig. 3). When the unconventional dielectrics were used, since they are not included in the software, kerosene data were selected. ^performance ^ Y ^e ^^e ^m (7) (8) (9) R 0,5 Fig. 2 Geometry of the micro-pocket Advances in Production Engineering & Management 14(3) 2019 347 Pellegrini, Ravasio Fig. 3 Implementation of dry-EDM For each pocket the energy consumption was measured using a watt-metro Christ ELEKTRONIK (CLM1000 Professional Plus) placed in way to include the whole power usage. When gaseous dielectrics were used, the dielectric unit was disabled to measure the actual electrical energy adsorbed by the machining. In the case of compressed air as dielectric, the energy consumed by the compressor was measured and included. At the end of each pocket, the electrode was cut using the wire EDM unit to restore the same initial electrode conditions for each test. For each milling, the EDM machine records the machining time, the electrode wear, the mean erosion speed, the eroded volume and the actual depth of the pocket. Table 2 reports the values of the coefficients used in the equations to calculate the sustainability index. Table 3 reports, for each dielectric used in the experimental investigation, the evaluation of the all the aspects taken into account for the determination of the penalty coefficient (A:;). This coefficient was used in the formula of the environmental sustainability (Se). For each aspect, a ranking index from 0 to 3 was evaluated. Coefficient kt is the ratio between the sum of all indexes and the worst possible score (all indexes equal to 3). Table 1 Properties of the dielectrics Type of dielectric Dynamic viscosity Density Dielectric Thermal Specific heat Dielectric [g/(m-s)] [g/dm3] rigidity conductivity [J/(g^K)] constant [kV/mm] [W/(m-K)] Kerosene 1.64 781 14-22 0.14-0.149 2.1-2.16 1.8 Water 0.92-1 1000 65-70 0.606-0.62 4.19 80.4 Vegetable oil 48.4 915-925 62-65 0.14-0.16 1.67 2.86 Air 0.019 1.205 3 0.016-0.026 1.005 1.000536 Oxygen 0.021 1.43 0.92-2.6 0.026 0.92 1.00049 Table 2 Values of the coefficients Kerosene H2O Vegetable oil Air Oxygen ce¡[€/kWh] 0.156 ce [€/mm] Brass: 0.024362; WC: 0.1054 Cad [€/l] 9.63 0.25 1.4 0 1.05 VdM 25 25 25 40 ¿d[h] 1000 1000 1000 33.33 caf [€] 117 Mh] 1000 cs [€/l] 0.215 1 1 1 1 cm[€/h] 40 Table 3 Determination of the penalty coefficient Kerosene Water Vegetable oil Compressed air Oxygen Fire hazard 3 0 0 0 2 Fumes production 3 3 3 0 0 Skin irritation 3 0 0 0 0 Toxic fumes 3 0 1 0 0 Dust production 0 0 0 3 3 Dielectric re-use 1 2 1 0 3 Dielectric dismantling 3 1 3 0 0 Filters dismantling 3 3 3 0 0 Total 19 9 11 3 8 ki 0.79 0.37 0.46 0.12 0.33 348 Advances in Production Engineering & Management 14(3) 2019 Evaluation of the sustainability of the micro-electrical discharge milling process 3. Results and discussion Figs. 4 and 5 show the erosion time and the volume of the electrode wear obtained milling AI-SI304 and Ti6Al4V using brass and WC electrodes varying the type of dielectric. The bars for brass electrode using gaseous dielectrics are omitted since these conditions did not permit to realize the test as previously underlined. In general, water as dielectric offers an optimal solution for all the tested conditions when the objective is to minimize the machining time. When brass electrode is used, vegetable oil is comparable to kerosene especially for AISI304 while using WC electrode there is a remarkable difference on the performance between the oil-based dielectrics: the machining occurs in a faster way using vegetable oil than kerosene. As far as gaseous dielectrics are involved, some interesting results are obtained: while compressed air does not represent a valid alternative, the oxygen is one of the best solutions to minimize the milling time. As regards the electrode wear, the gas dielectrics, especially the oxygen, minimize the wear. In general, the water as liquid dielectric obtains good results. It is hereby confirmed that vegetable oil is competitive with kerosene. Carbide electrodes show a lower electrode wear than brass [36]. Using these data, the developed Sustainability Index was calculated (Fig. 6). Several considerations can be made. First, the environmental impact process when brass electrode is adopted is lower than WC electrode. In fact, in three out of five sub-indexes of SI, machining time plays an important role and therefore the processes consuming lower time are more sustainable. Fixed the electrode, the dielectric that reveals to be the most sustainable is the water. In general, kerosene is less sustainable than the others liquid dielectrics. Vegetable oil is an appreciable dielectric in all the tested conditions. As regards gaseous dielectrics, compressive air gives worst results while oxygen is very interesting. On titanium sheets, oxygen is the best solution to minimize the sustainability index while on stainless steel gives good results. Anyway, this analysis on the global SI does not allow identifying the problematic issues of each experimented conditions. For this reason, Figs. 7 and 8 show the contribution of the five sustainability sub-indexes to the global SI. 1000 900 800 V 700 g 600 g> 500 I 400 S 300 200 100 0 Fig. 4 Machining time for AISI304 and Ti6Al4V using brass and WC electrode varying the dielectric 0,25 -0,2 -ST 0,15 - I 0,1 - 0,05 -0 - ■ Kerosene SH20 □ Vegetable Oil ■ Compressed Air ■ Oxygen ■ Kerosene SH20 □ Vegetable Oil ■ Compressed Air ■ Oxygen Fig. 5 Wear of brass and WC electrode when AISI304 and Ti6Al4V is machined varying the dielectric Advances in Production Engineering & Management 14(3) 2019 349 Pellegrini, Ravasio Using brass electrode, wear and environmental sub-indexes are especially relevant. The energetic component and that one related to the performance are almost constant for all the conditions. Regarding the others components, the dielectrics display the largest percentage variation, yet its contribution to SI is relatively small. The dielectric and environmental sustainability indexes are high for kerosene, medium for vegetable oil and low for water for both workpiece materials. Similar remarks are valid for electrode wear when AISI304 is machined, while electrode wear is almost the same for Ti6Al4V. There is a different situation when WC electrode is used. The energetic component results always small except for air as dielectric. For liquid dielectrics, each sub-index can be ranked as follows: high for kerosene, medium for vegetable oil and low for water. The main components are wear and environmental sub-indexes. Overall performance of air dielectric is poor while oxygen proves to be competitive with liquid dielectrics especially because it allows low electrode wear. Anyway, the critical aspects on the formation of the global SI varying workpiece and electrode material and dielectric type can be underlined though Figs. 6 and 7. In view of this analysis, it is possible to take actions aiming to reduce in general the sustainability index (and therefore to improve the sustainability level) focusing on the aspects causing more sustainability problems following a Pareto logic. i 0,9 0,8 0,7 0,6 0,5 0,4 0,3 0,2 0,1 0 ■ Kerosene SH20 □ Vegetable Oil ■ Compressed Air ■ Oxygen 1,41 k M Fig. 6 Sustainability index for AISI304 and Ti6Al4V using brass and WC electrode varying the dielectric Brass electrode 0,3 0,25 S °'2 0,15 0,1 0,05 0 H20 Vegetable Oil Kerosene AISI304 H20 TÍ6A14V VegOil I Senergy = Swear Senvironmental % Sperformance ■ Sdielectric Fig. 7 Composition of the sustainability index for AISI304 and Ti6Al4V using brass electrode varying the dielectric 350 Advances in Production Engineering & Management 14(3) 2019 Evaluation of the sustainability of the micro-electrical discharge milling process 1,8 1,6 1,4 1,2 1 v 0,8 55 0-6 0,4 0,2 0 WC electrode AISI304 T16A14V ■ Senergy Swear Senvironmental »■ Sperformance ■ Sdielectric Fig. 8 Composition of the sustainability index for AISI304 and Ti6A14V using WC electrode varying the dielectric A further comparison between working conditions (combinations of electrode material and dielectric) can be made by evaluating each single sub-index, normalizing the sum of all values corresponding to all working condition. The percentage distribution of each sub-indexes are reported in Figs. 9 and 10, showing a ranking of the experimental conditions for both work materials. Considering AISI304, in general the use of compressed air is not appreciable except for the indicator related to the dielectric sustainability. The combination kerosene as dielectric and tungsten carbide as electrode expends a lot of resources respect the others solutions. Oxygen as dielectric and WC electrode is the best solution for the sub-indexes regarding the energy consumption, the electrode wear and the environmental impact but need improvements for the performance and dielectric components. Water and vegetable oil always give good results. AISI304 ■ B_Kerosene □ B_VegOil s WC_H20 H WC Air H B_H20 ■ WC_Kerosene O WC_VegOil ■ WC_Oxygen Senergy I Swear Senvironmental Sperformance Sdielectric Fig. 9 Percentage distribution of sub-indexes on different experimental conditions in terms of electrode material and dielectric for AISI304 TÍ6A14V 70 60 # 50 40 30 20 10 0 Senergy ■ B_Kerosene B_H20 DB_VegOil ■ WC.Kerosene HWC_H20 WCJVegOil ■ WC.Air ■ WC.Oxygen Swear Senvironmental Sperformance Sdielectric Fig. 10 Percentage distribution of sub-indexes on different experimental conditions in terms of electrode material and dielectric for Ti6Al4V Advances in Production Engineering & Management 14(3) 2019 351 Pellegrini, Ravasio Also for Ti6Al4V, water and vegetable oil show good performance from the point of view of the sustainability. It is confirmed that compressed air is not a competitive dielectric. As regards the combination oxygen as dielectric and tungsten carbide as electrode, the critical aspect is only the dielectric sustainability while the others sub-indexes are very interesting. The advantage of the proposed index meets the requirements to be easily implemented in industrial applications. Anyway, the presented results in terms of the effect of the type of electrode and dielectric on the level of sustainability of the machining are influenced by the adopted parameters and by the choice of the values of the coefficients. The index could be improved taking into account other sustainability issues. For example, the pollution effects due to the contamination from dusts of both electrode and workpiece could be considered into the sub-index related the environmental. Moreover, other aspects related the quality of the machining could be taken into account such as roughness surface. The proposed index can be implemented in different technological situations such as micro-EDM drilling or WEDM. In fact, these different applications of the same technology have in common the same physical principle of material removal based on the erosion thorough electrical discharges between the workpiece and the electrode tool that occur in a dielectric fluid. The aspects taken into account for the elaboration of the sustainability index are in general common to other EDM processes and therefore the model can support similar works on others applications. In fact, the main factors related to sustainability of EDM processes are the energetic consumption, the electrode wear, the usage of the dielectric, the effect of the dielectric on the environmental and the process performance in terms of the probability of producing a non-conforming part. 4. Conclusion The micro-ED milling process was evaluated concerning the sustainability manufacturing. A global index, named Sustainability Index, taking into account the energetic consumption, the environmental impact, the dielectric consumption, the wear of the electrode and the machining performance (i.e. the scrapping/repairing rate) was developed. The index estimates the environmental impact in terms of both quantity of consumed resources and pollution effects created by the process. It was applied for an experimental case, in particular the execution of micro-pockets on stainless steel and titanium sheets using two types of electrode and five types of dielectric, both liquid and gaseous. For each workpiece material, the effects of both the electrode material and the type of dielectric on the sustainability process performance were analysed. In this way, for each condition in term of workpiece material/electrode material/dielectric the critical aspects related to the sustainability can be identified. Focusing on these aspects, actions of finding solutions minimizing the environmental impact of the process can be undertaken. Unusual dielectrics, such as vegetable oil and oxygen, proved to be valid substitutes to traditional ones under several viewpoints, including sustainability. The proposed index to measure the sustainability of micro-EDM milling process meets the requirements to be easily implemented in industrial applications. The index provides a tool that can assist the decision-making stage of the selection of the product and process conditions aiming the minimization of the environmental impact The obtained results can improve the knowledge of alternative dielectrics, not yet used in industrial applications. Finally, the aspects taken into account for the elaboration of the sustainability index are in general common to other EDM processes and therefore it can support similar works on others applications such as micro-EDM drilling or WEDM. Acknowledgements The authors wish to thank Dr. Fabio Caccia and Dr. Alessio Palomba from the University of Bergamo for their contribution in the experimental research. 352 Advances in Production Engineering & Management 14(3) 2019 Evaluation of the sustainability of the micro-electrical discharge milling process Reference [1] Singh, R.K., Murty, H.R., Gupta, S.K., Dikshit, A.K. (2009). An overview of sustainability assessment methodologies, Ecological Indicators, Vol. 9, No. 2, 189-212, doi: 10.1016/j.ecolind.2008.05.011. [2] Priarone, P. (2016). Quality-conscious optimization of energy consumption in a grinding process applying sustainability indicators, The International Journal of Advanced Manufacturing Technology, Vol. 86, No. 5-8, 21072117, doi: 10.1007/s00170-015-8310-9. [3] Hussain, S., Jahanzaib, M. (2018). Sustainable manufacturing - An overview and a conceptual framework for continuous transformation and competitiveness, Advances in Production Engineering & Management, Vol. 13, No. 3, 237-253, doi: 10.14743/apem2018.3.287. [4] Tan, X.C., Liu, F., Cao, H.J., Zhang, H. (2002). A decision-making framework model of cutting fluid selection for green manufacturing and a case study, Journal of Materials Processing Technology, Vol. 129, No. 1-3, 467-470, doi: 10.1016/s0924-0136(02)00614-3. [5] Valaki, J.B., Rathod, P.P., Khatri, B.C. (2015). Environmental impact, personnel health and operational safety aspects of electrical discharge machining: A review, Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture, Vol. 229, No. 9, 1481-1491, doi: 10.1177/0954405414543314. [6] Valaki, J.B., Rathod, P.P., Sankhavara, C.D. (2016). Investigations on technical feasibility of Jatropha curcas oil based bio dielectric fluid for sustainable electric discharge machining (EDM), Journal of Manufacturing Processes, Vol. 22, 151-160, doi: 10.1016/j.jmapro.2016.03.004. [7] Wang, X., Chen, L., Dan, B., Wang, F. (2018). Evaluation of EDM process for green manufacturing, The International Journal of Advanced Manufacturing Technology, Vol. 94, No. 1-4, 633-641, doi: 10.1007/s00170-017-0892- y. [8] Qin, Y. (2015). Micromanufacturing Engineering and Technology, 2nd Edition, Elsevier Inc., New York, USA, doi: 10.1016/C2013-0-19351-8. [9] Gutowski, T., Dahmus, J., Thiriez, A. (2006). Electrical energy requirements for manufacturing processes, In: Proceedings of 13th CIRP International Conference of Life Cycle Engineering, Leuven, Belgium, 623-627. [10] Kellens, K.; Renaldi; Dewulf, W.; Duflou, J.R. (2011). Preliminary environmental assessment of electrical discharge machining, In: Hesselbach, J., Herrmann, C. (eds.), Glocalized Solutions for Sustainability in Manufacturing, Springer, Berlin, Heidelberg, Germany, 377-382, doi: 10.1007/978-3-642-19692-8 65. [11] Kunieda, M., Lauwers, B., Rajurkar, K.P., Schumacher, B.M. (2005). Advancing EDM through fundamental insight into the process, CIRP Annals, Vol. 54, No. 2, 64-87, doi: 10.1016/s0007-8506(07)60020-1. [12] Das, M.K., Kumar, K., Barman, T.K., Sahoo, P. (2013). Optimization of surface roughness and MRR in EDM using WPCA, Procedia Engineering, Vol. 64, 446-455, doi: 10.1016/j.proeng.2013.09.118. [13] Gaikwad, A., Tiwari, A., Kumar, A., Singh, D. (2014). Effect of EDM parameters in obtaining maximum MRR and minimum EWR by machining SS 316 using copper electrode, International Journal of Mechanical Engineering and Technology, Vol. 5, No. 6, 101-109. [14] Abbas, N.M., Yusoff, N., Wahab, R.M. (2012). Electrical discharge machining (EDM): Practises in Malaysian industries and possible change towards green manufacturing, Procedia Engineering, Vol. 41, 1684-1688, doi: 10.1016/ j.proeng.2012.07.368. [15] Raju, L., Hiremath, S.S. (2016). A state-of-the-art review on micro electro-discharge machining, Procedia Technology, Vol. 25, 1281-1288, doi: 10.1016/j.protcy.2016.08.222. [16] Liu, Q., Zhang, Q., Zhang, M., Zhang, J. (2016). Review of size effects in micro electrical discharge machining, Precision Engineering, Vol. 44, 29-40, doi: 10.1016/j.precisioneng.2016.01.006. [17] Qian, J., Yang, F., Wang, J., Lauwers, B., Reynaerts, D. (2015). Material removal mechanism in low-energy micro-EDM process, CIRP Annals, Vol. 64, No. 1, 225-228, doi: 10.1016/j.cirp.2015.04.040. [18] Modica, F., Marrocco, V., Copani, G., Fassi, I. (2011). Sustainable micro-manufacturing of micro-components via micro electrical discharge machining, Sustainability, Vol. 3, No. 12, 2456-2469, doi: 10.3390/su3122456. [19] Bhuyan, R.K., Routara, B.C., Parida, A.K. (2015). Using entropy weight, OEC and fuzzy logic for optimizing the parameters during EDM of Al-24 % SiCP MMC, Advances in Production Engineering & Management, Vol. 10, No. 4, 217-227, doi: 10.14743/apem2015.4.204. [20] Rao, R.V., Rai, D.P., Ramkumar, J., Balic, J. (2016). A new multi-objective Jaya algorithm for optimization of modern machining processes, Advances in Production Engineering & Management, Vol. 11, No. 4, 271-286, doi: 10.14743/apem2016.4.226. [21] Singh, M., Ramkumar, J., Rao, R.V., Balic, J. (2019). Experimental investigation and multi-objective optimization of micro-wire electrical discharge machining of a titanium alloy using Jaya algorithm, Advances in Production Engineering & Management, Vol. 14, No. 2, 251-263, doi: 10.14743/apem2019.2.326. [22] Hourmand, M., Sarhan, A.A.D., Sayuti, M. (2017). Micro-electrode fabrication processes for micro-EDM drilling and milling: A state-of-the-art review, The International Journal of Advanced Manufacturing Technology, Vol. 91, No. 1-4, 1023-1056, doi: 10.1007/s00170-016-9671-4. [23] Daneshmand, S., Neyestanak, A.A.L., Monfared, V. (2016). Modelling and investigating the effect of input parameters on surface roughness in electrical discharge machining of CK45, Tehnicki Vjesnik - Technical Gazette, Vol. 23, No. 3, 725-730, doi: 10.17559/TV-20141024224809. [24] Chakraborty, S., Dey, V., Ghosh, S.K. (2015). A review on the use of dielectric fluids and their effects in electrical discharge machining characteristics, Precision Engineering, Vol. 40, 1-6, doi: 10.1016/j.precisioneng.2014.11. 003. Advances in Production Engineering & Management 14(3) 2019 353 Pellegrini, Ravasio [25] Niamat, M., Sarfraz, S., Aziz, H., Jahanzaib, M., Shehab, E., Ahmad, W., Hussain, S. (2017). Effect of different dielectrics on material removal rate, electrode wear rate and microstructure in EDM, Procedia C1RP, Vol. 60, 2-7, doi: 10.1016/j.procir.2017.02.023. [26] Srinivas Viswanth, V., Ramanujam, R., Rajyalakshmi, G. (2018). A review of research scope on sustainable and eco-friendly electrical discharge machining (E-EDM), Materials Today: Proceedings, Vol. 5, No. 5, Part 2, 1252512533, doi: 10.1016/j.matpr.2018.02.234. [27] Marashi, H., Jafarlou, D.M., Sarhan, A.A.D., Hamdi, M. (2016). State of the art in powder mixed dielectric for EDM applications, Precision Engineering, Vol. 46, 11-33, doi: 10.1016/j.precisioneng.2016.05.010. [28] Zhang, Y., Liu, Y., Shen, Y., Ji, R., Li, Z., Zheng, C. (2014). Investigation on the influence of the dielectrics on the material removal characteristics of EDM, Journal of Materials Processing Technology, Vol. 214, No. 5, 1052-1061, doi: 10.1016/j.jmatprotec.2013.12.012. [29] Shen, Y., Liu, Y., Dong, H., Zhang, K., Lv, L., Zhang, X., Zheng, C., Ji, R. (2017). Parameters optimization for sustainable machining of Ti6Al4V using a novel high-speed dry electrical discharge milling, The International Journal of Advanced Manufacturing Technology, Vol. 90, No. 9-12, 2733-2740, doi: 10.1007/s00170-016-9600-6. [30] Shirguppikar, S.S., Dabade, U.A. (2018). Experimental investigation of dry electric discharge machining (Dry EDM) process on bright mild steel, Materials Today: Proceedings, Vol. 5, No. 2, Part 2, 7595-7603, doi: 10.1016/ j.matpr.2017.11.432. [31] Singh, J., Sharma, R.V. (2017). Green EDM strategies to minimize environmental impact and improve process efficiency, Journal for Manufacturing Science and Production, Vol. 16, No. 4, 1-18, doi: 10.1515/jmsp-2016-0034. [32] Evertz, S., Dott, W., Eisentraeger, A. (2006). Electrical discharge machining: Occupational hygienic characterization using emission-based monitoring, International Journal of Hygiene and Environmental Health, Vol. 209, No. 5, 423-434, doi: 10.1016/j.ijheh.2006.04.005. [33] Marrocco, V., Modica, F., Fassi, I., Bianchi, G. (2017). Energetic consumption modelling of micro-EDM process, The International Journal of Advanced Manufacturing Technology, Vol. 93, No, 5-8, 1843-1852, doi: 10.1007/ s00170-017-0606-5. [34] Moldavska, A., Welo, T. (2017). The concept of sustainable manufacturing and its definition: A content-analysis based literature review, Journal of Cleaner Production, Vol. 166, 744-755, doi: 10.1016/j.jclepro.2017.08.006. [35] Valaki, J.B., Rathod, P.P. (2016). Assessment of operational feasibility of waste vegetable oil based bio-dielectric fluid for sustainable electric discharge machining (EDM), International Journal of Advanced Manufacturing Technology, Vol. 87, No. 5-8, 1509-1518, doi: 10.1007/s00170-015-7169-0. [36] D'Urso, G., Maccarini, G., Ravasio, C. (2016). Influence of electrode material in micro-EDM drilling of stainless steel and tungsten carbide, The International Journal of Advanced Manufacturing Technology, Vol. 85, No. 9-12, 2013-2025, doi: 10.1007/s00170-015-7010-9. 354 Advances in Production Engineering & Management 14(3) 2019 Advances in Production Engineering & Management Volume 14 | Number 3 | September 2019 | pp 355-366 https://doi.Org/10.14743/apem2019.3.333 ISSN 1854-6250 Journal home: apem-journal.org Original scientific paper A novel approximate dynamic programming approach for constrained equipment replacement problems: A case study Sadeghpour, H.a, Tavakoli, A.a*, Kazemi, M.a, Pooya, A.a aFerdowsi University of Mashhad, Faculty of Economics and Administrative Science, Mashhad, Iran A B S T R A C T A R T I C L E I N F O This paper presents a novel Approximate Dynamic Programming (ADP) approach to solve large-scale nonlinear constrained Equipment Replacement (ER) problems. Since ADP requires accurate estimations of states for future periods, a heuristic estimator based on data clustering was developed for the case study of this paper with small number of sampling periods. This ADP approach uses a Rollout Algorithm to formulate the problem in a Rolling horizon. The model was solved using Genetic Algorithm (GA). This framework was successfully applied for the decision making process of replacement/maintenance of 497 transformers in a power distribution company, which led to a significant reduction in the expected costs. The proposed framework possesses favourable features such as minimizing the effect of uncertainties in the state variables and measurement inaccuracies, which make the model robust and reliable. This work provides a novel general approach that can be employed for other industrial cases and operations research problems. © 2019 CPE, University of Maribor. All rights reserved. Keywords: Equipment replacement; Approximate dynamic programming; Rollout algorithm; State estimation; Genetic algorithm *Corresponding author: tavakoli-a@um.ac.ir (Tavakoli, A.) Article history: Received 7 May 2019 Revised 8 September 2019 Accepted 10 September 2019 1. Introduction Equipment replacement (ER) is an important decision process that nearly all industries are dealing with. ER optimization is a common topic in management science, and evolves constantly with the progress in operations research techniques. The literature available in this area focuses on decision making regarding maintenance or replacement of equipment over a limited or unlimited horizon, the examination of gradual changes in a technology, as well as the emergence of a new technology. ER problem has always been studied over the last century. In the early twentieth century, Taylor and Hoteling separately considered the cost of depreciation in ER calculations. In recent years, several studies have investigated the replacement of equipment such as transportation fleet [1], conveyor belts [2], medical equipment [3], reactor equipment (considering risk assessment) [4], heavy mining machinery [5], and information technology (IT) equipment [6]. In classical studies of ER, the goal is to find a policy to minimize discounted costs, while the interest rate and equipment costs usually remain constant. Annual costs of equipment operation and ownership are calculated during their lifetime, and an optimal life for the equipment is obtained by minimizing these costs. It can be clearly inferred from the previous works that the answer to a specific problem may considerably vary by changing the assumptions, where the answer usually changes from an optimal to a sub-optimal or even a non-optimal one. Future state variables are predicted from the previous and current ones, and the control parameters. State variables are evaluated based on measurements of a limited number of parameters with 355 Sadeghpour, Tavakoli, Kazemi, Pooya certain inaccuracies; therefore, the future state variables can only be predicted approximately. Leung and Tanchoco [7] proposed that some problems can be better handled by making an integrated decision about the equipment replacement. One of the most important factors which should be considered in making decisions, especially in ER, is to determine the prediction horizon, including limited, unlimited, and rolling horizons [8]. Fraser and Posey [9] presented a four-stage framework for analysing ER based on engineering economics, including determination of an alternative approach, prediction of the monetary flow for each approach, calculation of the present value of the monetary flow for each approach, and select of the solution method with the optimal present value. This method addresses ER on limited and unlimited horizons, as well as with and without considering technological changes. In modern economics literature, two categories of endogenous and exogenous factors are considered for economic fluctuations. In the 1960s, some economic theorists believed that economic fluctuations are similar to echoes, repeating over years with similar intensities and durations. Although this theory has been rejected in modern economics, Boucekkine, Germain, and Licandro [10] believed that this echo model can be modified and used for ER problems. They claimed that it can be shown that this echo is valid for ER. In addition, they proposed that profitability of different equipment with different technologies can be investigated by examination of possible solutions for a problem. However, this needs a huge amount of complicated calculation. Their study showed that the Dynamic Programming approach to ER is largely connected to the economic echoes model. One of the current problems in Dynamic Programming as well as in optimal control is to find a solution to integral equations to obtain an optimal policy. By applying an appropriate formulation, integral equations were developed for solving ER problems using Dynamic Programming [11], and later employed in several studies. Motamedi, Hadizadeh, and Peyghami [12] tried to find a numerical solution to the integral equations of [11]. They used the Adomian Decomposition Method to solve the equations, and presented a numerical example of ER to present the algorithm solution. Jacobsen [13] employed system dynamic methods for ER decisions. He first identified the subsystems and their components for his case study, and then estimated the future status of these subsystems using the existing data. The decision variable in his study was to repair or replace equipment. Two important features of a suitable model are its range of applicability for a particular problem, as well as its prevalence. By reviewing the previous studies on ER over the last 70-years, a common point can be clearly found, i.e., practical applications of most of the proposed models are not yet widespread. Therefore, it can be concluded that these models probably have not considered some practical factors with significant effects on decision-making process. According to [14], except a few cases of Stochastic Dynamic Programming there has been little progress in this area. Dynamic Programming solves a problem in successive steps, and adopts an optimal policy to satisfy the principle of optimality. An ER model seeks an optimal decision for preserving or replacement of equipment in consecutive time intervals; thus the Dynamic Programming method has been widely used in solving ER problems. Dynamic Programming was introduced in [15] and applied for ER by Bellman [16]. Dynamic Programming is in fact a general solution approach. Unlike linear or quadratic programming in which the structures of input data and analysis are quite clear, in solving a particular problem, the solution method should be adapted to the problem. Using the general structures proposed in Dynamic Programming, a unique solution method is established considering the main principle of Dynamic Programming - the principle of opti-mality [17]. On the other hand, unlike quadratic programing which can solve problems with many variables, the basic model of Dynamic Programming is only suitable for small-size problems. Increasing the number of variables usually increases the volume of computations, or in other word causes "curse of dimensionality". Many problems can be modelled and solved using Dynamic Programming. Depending on whether available information and variables are definite or random, various methods can be constructed to solve the problem. In the classical literature of Dynamic Programming, one can find well-known problems such as stagecoach in the shortest path, warehouse, distribution of effort, budgeting, Knapsack, and ER. The basic model of Dynam- 356 Advances in Production Engineering & Management 14(3) 2019 A novel approximate dynamic programming approach for constrained equipment replacement problems: A case study ic Programming is discrete, and mathematical principles and discrete control are employed to solve the model. Three main components of each Dynamic Programming model are state variable^), decision or control variable(s) and objective function. A trade-off always should be performed between the simplicity and possibility of a model analysis with the level of model details reflecting the real world conditions. To make a balance between them, two approaches can be applied: either the model can be simplified as much as possible, which is called the limited model, or it can be more complicated, which requires suboptimal methods to solve it Bertsekas [18] discussed that the efficiency of suboptimal methods is not less than limited models. In this study, sub-optimal methods were used. With the progress in operations research, different Dynamic Programming methods have been developed to build models very similar to real problems. A recent topic in Dynamic Programming has been the use of the ADP approach to solve large-scale and near-real-world problems. Some algorithms used in artificial intelligence such as queuing and game theory problems, and the optimal control are examples of modern problems closely related to Approximate Dynamic Programming [18]. In this study, the ADP approach has been used to formulate and solve an ER problem for the first time. First, the novel ADP framework to solve successive problems is explained, then the ER case study is described. Finally, it is discussed that the framework presented is able to provide optimal decisions for the ER problem studied. 2. Materials and methods Like other Dynamic Programming models, an ADP model tries to minimize (or maximize) an objective function considering constraints during a decision-making horizon. State variables for a limited or an unlimited horizon should be predicted based on control (decision) variables. The objective function is calculated from the state variables predicted, and then optimized using the optimal decision variables. Sub-optimal methods like heuristic and meta-heuristic algorithms are usually employed for optimization in ADP. A difference between definitive and approximate models is in their model state spaces, where an approximate model requires to predict the future state variables of a system using available data for the current state variables, which involves some uncertainties. Thus, before solving a model, the state variables should be estimated based on the control variables of the problem. Appropriate definitive data should be determined for the state space used in a model considering the relationship between data. This is usually done using trial and error, econometrics, data mining, heuristic, and dynamic neural programming methods to minimize the approximation error [18]. The conceptual ADP model adopted from [18] is shown in Fig. 1. Fig. 1 Conceptual ADP model used for ER problem 2.1 State estimation The first phase of ADP is the proper prediction of the system state. The applicability of multivariate methods (mostly used in econometrics, such as Auto Regressive Moving Average (ARMA) and Generalized Auto Regressive Conditional Heteroscedasticity (GARCH)) was examined, but appropriate results were not obtained due to low sampling periods of the case study. In similar studies, when the number of samples is high, but the number of measured periods is low, it has been suggested to use data mining techniques for state estimation [19]. Box and Meyer [20] stated that when the number of observations is much less than the number of samples, only a limited portion of data have the major effect on the prediction of samples, and called this situation as "Effect Sparsity". In this study, clustering of data has been used similar to the method proposed by [24]. This means that only those data that can provide the best approximation for Advances in Production Engineering & Management 14(3) 2019 357 Sadeghpour, Tavakoli, Kazemi, Pooya future states should be selected. Here, the aim is to apply an algorithm that minimizes the total error which originates from the estimation of existing data (Zz) and actual values (Zz). For this purpose, a heuristic algorithm with two estimators was developed, which acts as an intelligent filter to select data. This algorithm will be described in section 3.3 in details. 2.2 Rollout algorithm The model used for ADP in this study has been constructed based on "Rollout Algorithm", which is a sub-optimal control method for both definitive and stochastic systems. At each stage of decision-making process, the system is converted to a definite state by following specific standard steps, and then the Dynamic Programming or its equivalent optimal control problem is solved on a finite horizon from the current period (also called as rolling horizon). Subsequently, the first element of the decision parameters obtained is taken as the decision element of the current period and the rest are left out. In Rollout Algorithm, the objective function have been considered to be zero after the decision horizon [18]. In the second phase, based on the estimations obtained in the first phase, the problem is formulated and solved using Dynamic Programming. The system state was predicted considering the previous states and the applied decision variables. Although the decision space in this model is extended to several subsequent periods, the goal is to make a decision only in the current period. For the next periods, the process of modelling and problem solving is done again using more accurate inputs for the model. 2.3 Approximate dynamic programming model The conceptual model is represented as the mathematical model shown in Fig. 2. This a novel general approach that can be used in a variety of problems in Production Engineering and operations research when a regular decision-making process is required. The model is also applicable when the sampling period of measured data is low. q ^ Constraints Approximation Fig. 2 Schematic representation of the conceptual ADP model 3. Equipment replacement case study Khorasan-Razavi Electricity Distribution Company (KEDC) has the largest area of activity among other electricity distribution companies in Iran. KEDC distributes electricity at moderate (20 kV) and low (400 V) voltage levels in Khorasan Razavi province. Torbat-e Heydarieh Electricity office with over 103,000 consumers in 2018 is the third largest office of KEDC. Transformers are expensive equipment widely used in electricity distribution networks, and usually provide general or private power supply to one or several consumers. The case study here includes all 497 general pole-mount transformers in Torbat-e Heydarieh, where repairing and replacement of 358 Advances in Production Engineering & Management 14(3) 2019 A novel approximate dynamic programming approach for constrained equipment replacement problems: A case study these transformers impose a significant cost to the Company. These transformers have different capacities and ages. No structured method was previously used to make a schedule for maintenance or replacement of the transformers. This study used the transformers data available from 2013 to 2018 to build the ADP model, and identify transformers which possibly require replacement in 2019. For this purpose, it was necessary to choose suitable cost and objective functions, which were determined using a fuzzy method based on the available information and KEDC experts' opinion. 3.1 Formation of the database and normalization In this study, the Health Index (HI) of transformers is considered as the state variable. Three factors that affect the transformer HI are temperature difference, oil condition, and maximum load. To calculate the temperature difference, a thermo-vision camera was used to monitor the temperature at the outer surface of transformers. The difference of the hottest point of the transformer and the ambient temperature was recorded as the temperature difference. Insulator (oil) breakdown voltage tests were conducted for operating transformer to evaluate the oil condition. In this test, the temperature at which the oil loses its insulating properties was recorded. Finally, the maximum load is defined as the transformer load at peak times divided by its nominal capacity. These data were available for the years from 2013 to 2018. The data has been normalized using the max-min normalization method [21]. In this normalization method, if a factor is desirable to be higher, it is called a positive factor, and normalized as follows, - _ (1) X-max X-min however, if it is desirable to be lower, then is called as a negative factor, and normalized as follows, (2) where xs is the normalized value of x, and xmax and xmin are the maximum and minimum values of the data in the period, respectively. In this study, the maximum load and temperature difference are negative factors, while the oil condition is positive. 3.2 Calculation of the health index as a state variable HI for the transformer is defined as follows [22], HI = J J w J (3) where F is the number of the factors (F = 3 in this study), Wj is the weighting factor, and HIFj is the HI for each factor. The weighting factors were obtained using a Fuzzy AHP method with the help of the KEDC experts. The decision-making team were asked to prioritize the three HI factors considering their effects on the transformer health. If a transformer is replaced by a new one, the HI of the new transformer is considered to be one until the end of the decision horizon. 3.3 Creating the state estimator model This section explains how to organise the estimator for predicting the factors affecting the state of the model (HI of the transformers). The actual data (Zz) are available for six periods. As shown in Fig. 2, the data of each period in the first estimation step (Zz) are estimated using the linear regression model, and then used in the second estimator. In the second estimator, the data of similar transformers are averaged, and the final approximation data (Zz) are calculated and compared with the actual data (Zz). This procedure was carried out as follows: A. The period k in which the data are available is determined and repeated for k = 1 to K (K is number of previous periods with available data); B. For each transformer t, transformers are sorted from 1 to T (except for the period k) in a table based on the similarity of the data matrices (T is total number of transformers); Advances in Production Engineering & Management 14(3) 2019 SS9 Sadeghpour, Tavakoli, Kazemi, Pooya C. In each step, the following tasks are repeated for q = 1 to Q (q is the number of similar transformers and Q was determined 50): C.1. q similar transformers are selected (e.g., if q = 2, the transformer itself and the most similar transformer to that are selected). The average matrix is obtained by averaging over similar arrays. This 3D matrix has a dimension of (K— 1) *3*T (it consists of K — 1 periods in the first dimension, 3 factors in the second dimension, and T transformers in the third dimension). C.2. Using the average matrix of the previous step, the data for the period k are estimated by a linear regression model; C.3. Using a multivariable linear regression model for each factor, a linear set of equations are formed based on the other estimated factors in section C.2, and the data for the transformer t are estimated in the period k. D. The absolute differences of the estimated and actual values are calculated for all transformer factors in all periods, and their summation is considered as the model error. This task is done for different values of q similar transformers, and the error is calculated for each q. For each factor, the q with the minimum error is selected as the optimal number for the model estimators. In other words, using q similar transformers is recommended for the best state estimation. 3.4 Calculating the price of depreciated transformer In replacing transformers, the price of depreciated transformers should be considered. There are 10 different types of transformers with different capacities. To simplify the calculations, the price of a new 25 kVA transformer is considered one unit, and the price of other transformers is normalized to that. The value of a transformer used for I years is defined as follows, VT(l) = C0*e~l/A (4) where C0 is the price of a new transformer, A is the depreciation constant, and I is the age of the transformer. A was determined using fuzzy logic. Fuzzy logic uses fuzzy numbers instead of fixed and definite ones. This study employed the fuzzy logic method introduced by [23] and well described by [24]. The output of fuzzy calculations is a table which indicates the value of a depreciated transformer at different ages compared to a new transformer. For this purpose, a team of KEDC experts were asked to determine a minimum and a maximum price for transformers according to the age of transformers. These data were translated to fuzzy numbers, and tabulated in a data table. Finally, Eq. 4 was fitted to this data table. 3.5 Approximate dynamic programming formulation After finding the optimal estimator model, the objective function and the constraints should be rewritten in the form of the Rollout algorithm. This is done as follows (the total number of transformers is T, and each transformer and period are represented with indices of t and k, respectively). A. Using the estimator model proposed in the previous section and the data collected for the transformers, the factors for each transformer are predicted for the next period. B. Using the estimated factors and Eq. 3, HI values in each decision period (H/£) are obtained for all transformers in a form of a column matrix (HIk). By assembling this column matrices, a 2D matrix is formed, which shows the general state of the system when no control is applied to the system (no transformer is replaced during the decision horizon). C. utk is a zero-one control variable which indicates keeping or replacement of the transformer t in the period k; = 0 if a transformer is preserved, and utk = 1 if replacement is required. The state variable in the next period x£+1 is obtained using the transformer HI as follows, xfc+l = x0 *ufc + Pn(xfc-n+l ,xk-n+2 ,-■-, xfc)* C1 _ uk) (5) Eq. 5 shows that if it is decided to replace the transformer t, the situation is the same as the initial condition of the transformer installation (this transformer will not be replaced and its 360 Advances in Production Engineering & Management 14(3) 2019 A novel approximate dynamic programming approach for constrained equipment replacement problems: A case study HI remains constant until the end of the decision horizon). On the other hand, if it is decided to keep a transformer, the state value is estimated using the previous n data states as indicated in the steps A and B; Pn(xk-n+i, xk-n+2 '■■■' xk) presents this conditional value calculated by the state estimator. D. Based on the transformer state vector for each period, the costs of operation, maintenance and repair of each transformer, and a survey for transformers, the value of the transformer t is determined as, CK* = 0.15 * C0f *eo.o65*(i-(H/£)2)*i£ (6) where C0fis the price of a new transformer similar to the transformer t, and I£ is the age of the transformer t in the period k. E. CRk is the costs of purchasing new transformers and the replacement operation in the period k. The price of a new transformer is priori known, and the cost of the replacement operation is estimated to be 20 % of the price of a new transformer. Moreover, the depreciated transformer cost is subtracted from the replacement cost. This cost for each transformer t is calculated as, CRtk = 1.2 * Cq -VTQ) (7) where VT(l) is calculated using Eq. 4 as a function of the transformer age. F. The expected cost in each period is denoted as gk and is calculated as follows, T T = £ C4 x4 + ^ CK< x(1-u£) (8) 9k t=i t=i A constraint of the budget in each period bk is, T I I CRi x utk