ISSN 1854-6250 APEM journal Advances in Production Engineering & Management Volume 12 | Number 1 | March 2017 Published by PEI 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 Production Engineering Institute (PEI), 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 Tomaz Irgolic deski@apem-journal.org Website Technical Editor Lucija Brezocnik lucija.brezocnik@um.si 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 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. Production Engineering Institute 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 © 2017 PEI, University of Maribor. All rights reserved. Advances in Production Engineering & Management is indexed and abstracted in the WEB OF SCIENCE (maintained by THOMSON REUTERS): 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 Production Engineering Institute (PEI) Advances in Production Engineering & Management Volume 12 | Number 1 | March 2017 | pp 1-100 Contents Scope and topics 4 Solving dual flexible job-shop scheduling problem using a Bat Algorithm 5 Xu, H.; Bao, Z.R.; Zhang, T. Performance modelling based on value analysis for improving product development 17 process architecture Yin, F.P.; Gao, Q.; Ji, X. The impact of information and communication technologies (ICT) on agility, operating, and 29 economical performance of supply chain García-Alcaraz, J.L.; Maldonado-Macías, A.A.; Alor-Hernández, G.; Sánchez-Ramírez, C. Fuzzy Delphi and hybrid AH-MATEL integration for monitoring of paint utilization 41 Kumar, A.; Mussada, E.K.; Ashif, M.; Tyagi, D.; Srivastava, A.K. Reconsidering production coordination: A principal-agent theory-based analysis 51 Gong, D.; Tang, M.; Liu, S.; Li, Q. A genetic regulatory network-based sequencing method for mixed-model assembly lines 62 Lv, Y.; Zhang, J.; Qin, W. Influence of the production fluctuation on the process energy intensity in iron and 75 steel industry Chen, D.; Lu, B.; Chen, G.; Yu, W. Automatic high resolution measurement set-up for calibrating precise line scales 88 Klobucar, R.; Acko, B. Calendar of events 97 Notes for contributors 99 Journal homepage: apem-journal.org ISSN 1854-6250 (print) ISSN 1855-6531 (on-line) ©2017 PEI, University of Maribor. All rights reserved. 3 Scope and topics Advances in Production Engineering & Management (APEM journal) is an interdisciplinary refer-eed international academic journal published quarterly by the Production Engineering Institute 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 Computer-Integrate d M anufacturing Cutting and Forming Processes Decision Support Systems 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 Machine Learning in Production Machine Tools Machining Systems Manufacturing Systems Materials Science, Multidisciplinary Mechanical Engineering Mechatronics Metrology 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 4 APEM jowatal Advances in Production Engineering & Management Volume 12 | Number 1 | March 2017 | pp 5-16 https://doi.Org/10.14743/apem2017.1.235 ISSN 1854-6250 Journal home: apem-journal.org Original scientific paper Solving dual flexible job-shop scheduling problem using a Bat Algorithm Xu, H.a*, Bao, Z.R.a, Zhang, T.a aSchool of Internet of Things Engineering, Jiangnan University, Wuxi, China A B S T R A C T A R T I C L E I N F O For the flexible job-shop scheduling problem with machine selection flexibility and process sequence flexibility in process design, types and characteristic of machine selection and process sequence flexibility are analyzed. The mathematical model of dual flexible job-shop scheduling problem is established, and an improved bat algorithm is proposed. For purpose of expressing the relationship effectively between the process and the bat population, a new method of encoding strategy based on dual flexibility degree is proposed. The crossover and mutation operation are designed to strengthen the searching ability of the algorithm. For purpose of overcoming the shortcomings of the fixed parameters in bat algorithm, the value of the inertia weight was adjusted, and a linear decreasing inertia weight strategy was proposed. We carried out experiments on actual examples, it can be seen from the experimental results that the robustness and optimization ability of the algorithm we proposed are better than Genetic Algorithm (GA) and Discrete Particle Swarm Optimization algorithm (DPSO). This shows that the proposed algorithm is more excellent in solving the flexible job-shop scheduling problem, and it is an efficient scheduling algorithm. © 2017 PEI, University of Maribor. All rights reserved. Keywords: Flexible job-shop scheduling Optimization Process sequence flexibility Machine selection flexibility Bat algorithm Genetic algorithm Particle swarm optimization *Corresponding author: joanxh2003@163.com (Xu, H.) Article history: Received 16 November 2016 Revised 13 February 2017 Accepted 15 February 2017 1. Introduction Dual Flexible Job-shop Scheduling Problem (DFJSP) based on machine selection flexibility and process sequence flexibility is an extension of the classical Job-shop Scheduling Problem (JSP) and the Flexible Job-shop Scheduling Problem (FJSP). It overcomes the problems of the fixed step sequence and uniqueness of machine selection, and increases the flexibility of scheduling problem, making it closer to an actual production environment. DFJSP needs to consider not only the processing sequence of the work, but also the problems of assigning operations to machines, which are a more complicated NP-hard problem [1]. Therefore, research on DFJSP has theoretical significance and application value. At present, researches on FJSP are mainly focused on two aspects. The first involves assigning operations to machines. Yuan et al. [2] proposed a hybrid harmony search algorithm for solving FJSP, they converted the continuous vector to the discrete vector and introduced heuristic and random strategies for a resentful initialization scheme. Zhao et al. [3-4] proposed a hybrid genetic algorithm to solve FJSP. Luan et al. [5] proposed a new genetic algorithm to solve FJSP, in which they optimize the performance by minimizing the data sizes of the constrained model and reduced the time and space complexity of GA. The second involves processing sequence flexibility. A model is defined by Saygin[6] , which combines flexible process design and production 5 Xu, Bao, Zhang scheduling. Huang et al. [7] proposed an improved genetic algorithm for solving JSP based on process sequence flexibility. Ba et al. [8] proposed a new model of multi-resource flexible job shop scheduling problem, and designed a new genetic algorithm for solving this problem. Moreover, Modrak et al. [9] proposed an algorithm that can convert a multi-machines problem to a two-machines problem, and experiments show that the algorithm is effective in solving n-jobs and m-machine problems. Bat Algorithm (BA) is a materialistic optimization algorithm developed by Yang [10] in 2010. Since its proposed the algorithm has caught the attention of scholars in different fields. It has been used in the study of power system stability by Ali [11]. Shi et al. [12] has applied BA to WSN position. Chen et al. [13-14] proposed a mind evolutionary bat algorithm which is applied to feature selection of mixed gases infrared spectrum. Besides its relevance to practical application BA is applied to the function optimization problem [15], pattern recognition [16] and optimization of engineering problems [17]. In area of workshop scheduling, there are a lot of articles concerning the influencing factors since the BA was presented. Marichelvam el al. [18-19] has applied the BA to the hybrid flow shop scheduling problem (FSP) in 2013, and it shows BA was more effective than genetic algorithm and particle swarm optimization in solving this problem through the simulation results. Luo et al. [20] proposed a discrete BA solution for permutation flow shop scheduling problem (PFSP) in 2014 and it also shows that BA was effective in solving PFSP through the simulation results. Zhang et al. [21] proposed an improved BA for solving PFSP, the simulation results show that the improved BA is feasible and effective. La et al. [22] proposed a hybrid BA to solve the two stages hybrid FSP and the result shows that the hybrid BA is better than the classical BA. Be seen from the above, the studies of FJSP are primarily focused on the single FJSP. DFJSP is less researched. Although BA is applied to various fields, few investigators utilize it to solve DFJSP. Therefore, this paper establishes a model for DFJSP and proposes an improved bat algorithm for solving it. 2. Definition and formalization problem Assumptions of notations for DFJSP are as follows: • Let J = Ji, 1 < i is denoted as the flexible process sequence span of job /¿. Flexible operation is classified into two types as follows: (1) Partial flexible operation: An operation which belongs to a flexible process sequence span can be inserted into any arbitrary position of the span. (2) Total flexible operation: An operation can be inserted into any arbitrary position of the process sequence. Machine selection flexibility The machine selection flexible scheduling problems can be classified into two main groups according to the relation between set M^ and set M. (1) Total machine selection flexible scheduling problem: Each operation can be processed on any machine among the set Mtj, where Mij = M. (2) Partial machine selection flexible scheduling problem: Each operation can be processed on any machine among the set M^, where M^ c M. The partial machine selection flexible scheduling problem suits better than the total machine selection flexible scheduling problem in the practical manufacturing environments. But compared to the total machine selection flexible scheduling problem, the partial machine selection flexible scheduling problem has the disadvantages of great search space, great computation and difficulties solution. Table 1 An example of processing time table of partial machine selection flexible scheduling problem Job Operation M1 M2 M3 M4 On 2 7 — 6 A 0l2 6 — 4 5 0l3 — 3 2 4 02i 3 1 6 3 h o22 1 3 — 3 Table 2 An example of processing time table of total machine selection flexible scheduling problem Job Operation M1 M2 m3 M4 On 2 7 2 6 A 0l2 6 3 4 5 0l3 4 3 2 4 02i 3 1 6 3 h o22 1 3 4 3 Examples of partial machine selection flexible scheduling problem and a total machine selection flexible scheduling problem are shown in Table 1 and Table 2 respectively. Processing time of each operation on the corresponding machine is indicated in the table. In Table 1, the tag "—" means that a machine cannot execute the corresponding operation. Assumptions used throughout the DFJSP are as follows: • In flexible process sequence span, each operation has the same priority. • Each operation, once started, cannot be interrupted. • Each machine can process only one job at the same time. • Arrival time of a job is included in the processing time. • When machines are available for re-scheduling when the corresponding operations are completed, machines can choose to stop or no-load running. The task of DFJSP is to seek an appropriate schedule which cost minimum time to complete all operations. The makespan Tmax can be calculated by the formula: Tmax = max{7^ax| j = 1,2,..., m] (1) where T^ax is time of completion of last job on machine j. Advances in Production Engineering & Management 12(1) 2017 7 Xu, Bao, Zhang 3. The improved bat algorithm To effectively avoid the bat algorithm from prematurity and to improve the global search capability and search precision, in this paper the following improvements of BA are made: (1) In order to express the relationship between the process flexibility, process sequence and bat population, a new coding strategy based on dual flexibility is proposed. (2) To enhance the ability of neighborhood search, the corresponding operations like crossover and mutation are designed. (3) A linear decreasing inertia weight strategy is adopted to effectively control the local and global search capability of BA during the optimization procedure. 3.1 Coding strategies Coding is to abstract and specify the research problem, and then to develop mathematical model. Finally, it realizes the mapping between the solution space of feasible solution and the exploration space of BA. This is the key step in BA solving, and also the primary problem to be solved. Classical JSP only needs to code based on the process sequence. But DFJSP need not only to code based on process sequence, but also need to select the corresponding machine for each operation. Therefore, the coding strategy needs to be done in the following three points: (1) The coding strategy should reflect machine selection flexibility and process sequence flexibility. (2) The coding strategy should show the sequence of operations for each job. (3) It has to show the corresponding processing machine that each job need for each operation. Thus, we design three coding strategy based on above requirements: A coding strategy base on the sequence In the bat algorithm, all processes are required to be added to the code, present with vector Xx = (x11,x12,..,x1^) , the length is A, among those vector elements, x11 indicates the first process of the first job, x12 indicates the second process of the first job. By that analogy, the total number of t is the operation of the i-th job characters of y'-th job. In vector X1, the order of the ith corresponds to the order of processing sequence of each process. The advantages of this coding strategy are the high flexibility and always generate the feasible schedule after an exchange of the processing sequence. A coding strategy base on the process sequence flexibility In the DFJSP, after the coding based on sequence, it is needed to mark and determine the priority of each job with flexible process sequence span, here we use vector Xi = (x2i,x22,",X2a) to present, the length is X. Among them, the default initialization for x2i is 0 or 1, 0 indicates that the process sequence is not flexible, to 1, indicates the opposite. A coding strategy base on the machine selection flexibility In the bat algorithm, we use vector X3 = (x31,x32,...,x3%) to present machine selection. The length is A. Among them, each variable is a positive integer which represents the position in the selected machine set. The advantage of this coding strategy is to ensure that generate the feasible schedule after the operation, and it can be applied to either totally FJSP or partially FJSP. 3.2 Bat algorithm and its improvement Assume search space is a d-dimensional space, v^t — 1) and X;(t — 1) denote the velocity and location of bat i at the time step t-1. The location xt (t) and velocity Vi (t) at the time step t are as follows: where Ft is the update frequency of bat i, Fmin and Fmax represent the minimum and maximum value of the update frequency. ft e [0,1] and £ e [—1,1] are random numbers. Here x* is the cur- (2) (3) (4) (5) (6) (7) At{i) = aAt~1(i) ßt(i) = ß0(i)[l-exp(-7(i-l))] 8 Advances in Production Engineering & Management 12(1) 2017 Solving dual flexible job-shop scheduling problem using a Bat Algorithm rent global optimal solution, xoM is an optimal solution from a random selection of the optimal solution set, is the average loudness at the time step t — 1. For any 0 < a < 1 (ffEfi) or y > 0 (y E R), we have, At(i) ^0,Rt(j) as t^œ (8) The standard bat algorithm flow is presented in Fig. 1. Begin Step 1. Step 2. Step 3. Step 4. Step 5. Step 6. Step 7. Step 8. Step 9. Step 10. Step 11. Step 12. Step 13. Step 14. Step 15. Step 16. Step 17. Step 18. End Initialize parameters such as bat population xt, velocity vt, loudness A(i), pulse rates R(i) and so on Define pulse frequency Fj Calculated fitness value fitmin and the current global optimal solution x* While (t < Max number iterations) For (i = 1 to pop size) Generate new solutions by adjusting frequency, and updating velocities and locations via Eq. 2 to Eq. 4 If (rand>fl(i)) Select a solution among the optimal solution set Generate a new solution via Eq. 5 end if Calculated fitness value fitnew If (fitnew means to mutate the selection of by value from Mtj. Updating process of position and velocity in bat algorithm is similar to the updating process of PSO algorithm [24]. The inertia weightw is introduced to balance the local searching and the global searching via equation (9) [25]. In order to enhance the algorithm's ability of global searching at prophase of evolution, the weight exponential decline strategy is applied as shown in equation (10). Vi (t) = wvi(t- 1) + (xt (t-1)-x*)Fi (9) w = T-^- (wmax -wmin) + wmin (10) Where, t is the current number of iterations and T is maximum number of iterations; wmax and wmin are the maximum and minimum values of the inertia weight. To summarize, the basic steps of the improved bat algorithm for solving DFJSP can be described as follows in Fig 2. Step 1. Initialize parameters, include the size of bat population PopSize, loudness A(i), pulse rates R (¿), maximum inertia weight wmax, minimum inertia weight wmin and so on Step 2. The coding sequence is generated based on these three coding strategies Step 3. Calculate fitness value Step 3.1 Load the information of all jobs into PJ and TRP Step 3.2 Every job is processed in the beginning from the first operation by PJ. Firstly, judge by RS whether the machine which the first operation needing is processed is used by another operation. If not, then the job is into the processing status. Meanwhile, record the corresponding numbers of the job and the operation. Otherwise, the job is flagged as waiting for processing in PJ. Step 3.3 The first job starts to be processed in PJ and TRP, which judge by TRP is being processed. If it is and the rest time of processing is greater than zero, the processing of the job is going on. If not, judge whether all operations of the job are finished. If the job is accomplished, make the machines free by RS. Otherwise, add the information of the next operation into PJ and TRP. Step 3.4 If all jobs are already processed by PJ, go to Step 3.5; otherwise, go to Step 3.2 Step 3.5 Output fitness value Step 4. If the termination criterion is reached, go to Step 10; otherwise, go to Step 5 Step 5. Generate new solutions by adjusting frequency, and updating velocities and locations via Eqs. 2, 9 and 10 Step 6. If rand>fl(i), select a solution from optimal solution set and generate a local solution xnew around the selected optimal solution via Eq. 5 Step 7. Calculate fitness value fitnew of corresponding xnew Step 8. If fitnew , <5,6> h 3 — h 5 <3,5> h 3 <2,3> Js 6 <3,4> h 4 — The tag "—" means that the job does not have a flexible process span. 12 Advances in Production Engineering & Management 12(1) 2017 Solving dual flexible job-shop scheduling problem using a Bat Algorithm Table 5 Data of process sequence flexibility and processing time (min) Job Operation — M2 M4 Mc, — m7 mr On 12 — — 9 14 20 Ol2 18 — 19 — 15 11 — A Ol3 — 12 14 9 — — 17 — — 014 — 11 — 9 — — 12 Ois 15 — — 8 — — — — 18 0l6 12 — 9 — 11 — — o2i — — 12 — 19 14 — — — h o22 8 — 9 — 11 — — 15 — 023 16 7 — 9 — — 031 — — — 11 10 — — — 13 — 032 — — 12 18 — — 14 — h o33 9 — — — 15 7 12 034 12 — — 15 — 7 — 9 035 3 — — 4 — 8 — — 041 — 19 — — — 7 — 13 — — J* 042 8 — 11 — 16 — 043 9 11 — 8 — — 18 Osl 6 — — 12 — — 14 9 — 052 — — — 22 — 12 17 Js 0S3 — 18 — — — 11 — — 9 054 9 — 12 — — — — — — 7 — 055 11 — — 9 14 — 056 8 — — 12 6 — — 9 061 — 11 — 17 — 18 h 062 5 — — 12 — — — — 7 9 063 11 — — — — 8 13 — 7 — 0M 19 13 7 15 4.3 Experimental results and comparisons This paper used the improved bat algorithm, discrete particle swarm optimization [26] (DPSO) and genetic algorithm [27] (GA) to solve the DFJSP. And the experimental results are compared and analyzed. To obtain meaningful results, we run each algorithm fifty times on the above instance. The experimental results of these three kinds of algorithms are shown in Table 6 and Table 7. The optimum solution, average solution and standard deviation of three algorithms in 50 experiments are shown in Table 6. The optimum solution obtained by this paper is 55, which is obviously better than 65 obtained by GA and 62 obtained by DPSO. Furthermore, the average value and standard deviation are also better than both GA and DPSO. In a conclusion, the algorithm in this paper is obviously better than GA and DPSO in searching performance. Table 6 Makespan comparison between three algorithms Algorithm Optimum solution Average solution Standard deviation GA 65 70.72 4.25 DPSO 62 65.56 3.15 Improved bat algorithm 55 57.32 2.24 Table 7 Distribution of 50 calculation results ———___Interval Interval Algorithm —■—____ [55,60) [60,65) [65,70) [70,75) [75,80) GA 0 0 23 14 13 DPSO 0 23 19 8 0 Improved bat algorithm 40 10 0 0 0 The distribution of experiments results in each interval for running 50 times is presented in Table 7. It is clear from the table, the results of the improved bat algorithm are comparatively concentrated, which are mainly in the range of [55, 60], whereas the results of GA and DPSO algorithm are relatively distributed. Thus, the improved bat algorithm in terms of solving DFJSP Advances in Production Engineering & Management 12(1) 2017 13 Xu, Bao, Zhang is more robust than GA and DPSO algorithm. Figs. 3, 4, and 5 shows the Gantt charts of optimal scheduling concerning three kinds of algorithms. ms m7 Mb m5 m4 m3 m2 mi 3-1 5-3 5-4 _13_1S_ 1-2 3-2 5-2 6-3 3-4 4-1 1-4 3-3 1-1 1 1 1-3 1 1 3-5 1 1-5 1 6-4 1 9 20 29 34 38 46 53 2-1 1 6-1 1 4-2 1 1 1-6 1 12 23 31 46 55 1 4-3 1 1 2-3 1 1 5-5 7 18 20 27 34 4 5 5-1 1 1 2-2 1 1 6-2 1 5-6 1 5 10 15 20 25 30 35 40 45 50 55 Fig. 3 The Gantt chart of optimal scheduling of the improved bat algorithm 60 27 34 20 9 36 38 6 28 18 29 45 7 52 6 12 20 23 28 45 53 m8 m7 mb m5 m4 m3 m2 m1 3-1 5-3 _13_12- 1-2 I I_3-3_ 20_25_ 5-2_I 6-3 I_6-4_I 2-3 I 3-4 50 2-1 I 2-2 I 4-1 I 4-2 I I 5-6 51 1-1 I I 1-3 I I 1-5 I l~3-5~l 48_57 61 6-1 I 3-2 I I 5-4 I I 1-6 1-4__5-5__4-3 5-1 I I 6-2 6 11 16 5 10 15 20 25 30 35 40 45 50 55 60 65 Fig. 4 The Gantt chart of optimal scheduling of DPSO algorithm 27 37 11 13 25 27 39 57 51 32 29 40 m8 m7 mb m5 m4 m3 m2 mi 5-1 I 6-1 I 5-3 I 5-4 I 3-4 5-2_ I 6-2 1-2 3-1 I I 4-2 I I 1-4 H I 5-6 I 47_ œ_ 62 I 1-5 I 3-5 I 55_53_ 1-1 3-2 1 9 10 28 2-1 1 1 12 24 4-1 1 1 2-3 1 19 20 27 1 2 -2 1 1 12 20 28 5 10 15 20 25 1-3_I I 1-6 I-33--5^ I 6-3 I 5-5 \ _34_45 _56_ 3-3 I 4-3 I 6-4 65 -► Fig. 5 The Gantt chart of optimal scheduling of GA 10_ 27 52 36 34 26 27 64 46 37 14 Advances in Production Engineering & Management 12(1) 2017 Solving dual flexible job-shop scheduling problem using a Bat Algorithm 5. Conclusion According to dual flexible job shop scheduling problems, we established a mathematical model based on process sequence flexibility and machine selection flexibility, and an improved bat algorithm is presented to solve it. The premise of BA used in DFJSP is to realize the mapping between operations and bat populations. Therefore, we proposed a dual flexible encoding strategy. In order to intensify neighborhood searching ability of the algorithm, several operations are designed such as crossover and mutation. Furthermore, a linear decreasing inertia weight strategy is put forward to effectively avoid the algorithm of premature convergence and to intensify the global searching ability and the precision of the algorithm. Aiming at solving actual job-shop scheduling problems, we set up the scheduling model and designed the detailed algorithm. Experimental results indicate that the improved bat algorithm for solving DFJSP is feasible and effective, which provides a new way to solve such problems. It is a further research direction to use the improved algorithm to solve multi-objective DFJSP and construct new dispatching rules. Acknowledgement The research has been supported by Natural Science Foundation of Jiangsu Province (Grant No. BK20140165) and China Scholarship Council Sponsored (Grant No. 201308320030). References [1] Rock, H. (1984). The three-machine no-wait flow shop problem is NP-complete, Journal of the ACM, Vol. 31, No. 2, 336-345, doi: 10.1145/62.65. [2] Yuan, Y., Xu, H., Yang, J. (2013). A hybrid harmony search algorithm for the flexible job shop scheduling problem, Applied Soft Computing, Vol. 13, No. 7, 3259-3272, doi: 10/1016/j.asoc.2013.02.013. [3] Zhao, S.-K., Fang, S.-L., Gu, X.-J. (2013). Genetic algorithm with new initialization mechanism for flexible job shop scheduling, Journal of Zhejiang University (Engineering Science), Vol. 47, No. 6, 1022-1030, doi: 10.3785/j.issn. 1008-973X.2013.06.013. [4] Zhao, S.-K., Fang, S.-L., Gu, X.-J. (2014). Machine selection and FJSP solution based on limit scheduling completion time minimization, Computer Integrated Manufacturing Systems, Vol. 20, No. 4, 854-865, doi: 10.13196/j.cims. 2014.04.zhaoshikui.0854.12.20140416. [5] Luan, F., Wang, W., Fu, W.P., Bao, Y.T., Ren, G.C., Wang, J., Deng, M.M. (2014). FJSP solving by improved GA based on PST hierarchy structure, Computer Integrated Manufacturing Systems, Vol. 20, No. 10, 2494-2501. [6] Saygin, C., Kilic, S.E. (1999). Integrating flexible process plans with scheduling in flexible manufacturing systems, The International Journal of Advanced Manufacturing Technology, Vol. 15, No. 4, 268-280, doi: 10.1007/ s001700050066. [7] Huang, X.W., Zhao, X.Y., Ma, X.L. (2014). An improved genetic algorithm for job-shop scheduling problem with process sequence flexibility, International Journal of Simulation Modelling, Vol. 13, No. 4, 510-522, doi: 10.2507/ ijsimm13(4)co20. [8] Ba, L., Li, Y., Yang, M.S., Gao, X.Q., Liu, Y. (2016). Modelling and simulation of a multi-resource flexible job-shop scheduling, International Journal of Simulation Modelling, Vol. 15, No. 1, 157-169, doi: 10.2507/IJSIMM15(1)C03. [9] Modrak, V. Pandian, R.S. (2010). Flow shop scheduling algorithm to minimize completion time for n-jobs m-machines problem, Tehnicki vjesnik -Technical Gazette, Vol. 17, No. 3, 273-278. [10] Yang, X.-S. (2010). A new materialistic bat-inspired algorithm, Nature Inspired Cooperative Strategies for Optimization, Vol. 284, 65-74, doi: 10.1007/978-3-642-12538-6 6. [11] Ali, E.S. (2014). Optimization of power system stabilizers using bat search algorithm, International Journal of Electrical Power & Energy Systems, Vol. 61, 683-690, doi: 10.1016/i.iiepes.2014.04.007. [12] Shi, H., Wang, W., Li, Y., Lu, L. (2015). Bat algorithm based on Levy flight feature and its localization application in WSN, Chinese Journal of Sensors and Actuators, Vol. 6, 888-894, doi: 10.3969/j.issn.1004-1699.2015.06.019. [13] Chen, Y., Wang, Z., Wang, Z. (2014). Feature selection of infrared spectrum based on improved bat algorithm, Infrared and Laser Engineering, Vol. 43, No. 8, 2715-2721, doi: 10.3969/j.issn.1007-2276.2014.08.054. [14] Chen, Y., Wang, Z., Wang, Z. (2015). Mind evolutionary bat algorithm and its application to feature selection of mixed gases infrared spectrum, Infrared and Laser Engineering, Vol. 3, 845-851, doi: 10.3969/j.issn.1007-2276. 2015.03.010. [15] Yang, X.-S. (2011). Bat algorithm for multi-objective optimization, International Journal of Bio-Inspired Computation, Vol. 3, No. 5, 267-274, doi: 10.1504/IIBIC.2011.042259. [16] Komarasamy, G., Wahi, A. (2012). An optimized k-means clustering technique using bat algorithm, European Journal of Scientific Research, Vol. 84, No. 2, 263-273. [17] Yang, X.-S., Gandomi, A.H. (2012). Bat algorithm: A novel approach for global engineering optimization, Engineering Compotations, Vol. 29, No. 5, 464-483, doi: 10.1108/02644401211235834. Advances in Production Engineering & Management 12(1) 2017 15 Xu, Bao, Zhang [18] Marichelvam, M.K., Prabaharan, T. (2012). A bat algorithm for realistic hybrid flowshop scheduling problems to minimize makespan and mean flow time, ICTACTJournal on Soft Computing, Vol. 3, No. 1, 428-433, doi: 10.21917 /ijsc.2012.0066. [19] Marichelvam, M.K., Prabaharan, T., Yang, X.-S., Geetha, M. (2013). Solving hybrid flow shop scheduling problems using bat algorithm, International Journal of Logistics Economics and Globalization, Vol. 5, No. 1, 15-29, doi: 10.1504 /IILEG.2013.054428. [20] Luo, Q., Zhou, Y., Xie, J., Ma, M., Li, L. (2014). Discrete bat algorithm for optimal problem of permutation flow shop scheduling, The Scientific World Journal, Vol. 2014, 1-15, doi: 10.1155/2014/630280. [21] Zhang, J.J., Li, Y.G. (2014). An improved bat algorithm and its application in permutation flow shop scheduling problem, Advanced Materials Research, Vol. 1049-1050, 1359-1362, doi: 10.4028/www.scientific.net/amr.1049-1050.1359. [22] Dekhici, L., Belkadi, K. (2015). A bat algorithm with generalized walk for the two-stage hybrid flow shop problem, International Journal of Decision Support System Technology, Vol. 7, No. 3, 1-16, doi: 10.4018/ijdsst.2015070101. [23] Mirjalili, S., Mirjalili, S.M., Yang, X.-S. (2014). Binary bat algorithm, Neural Computing and Applications, Vol. 25, No. 3, 663-681, doi: 10.1007/s00521-013-1525-5. [24] Bai, J., Gong, Y.-G., Wang, N.-S., Tang, D.-B. (2010). Multi-objective flexible job shop scheduling with lot-splitting, Computer Integrated Manufacturing Systems, Vol. 16, No. 2, 396-403. [25] Zhao, S. (2015). Bilevel neighborhood search hybrid algorithm for the flexible job shop scheduling problem, Journal of Mechanical Engineering, Vol. 51, No. 14, 175-184. [26] Xu, H., Zhang, T. (2015). Improved discrete particle swarm algorithm for solving flexible flow shop scheduling problem, Journal of Computer Application, Vol. 35, No. 5, 1342-1347, doi: 10.11772/j.issn.1001-9081.2015.05. 1342. [27] Cui, J.S., Li, T.K., Zhang, W.X. (2005). Hybrid flow shop scheduling model and its genetic algorithm, Journal of University of Science and Technology Beijing, Vol. 27, No. 5, 623-626, doi: 10.3321/j.issn:1001-053X.2005.05.027. 16 Advances in Production Engineering & Management 12(1) 2017 Advances in Production Engineering & Management Volume 12 | Number 1 | March 2017 | pp 17-28 https://doi.Org/10.14743/apem2017.1.236 ISSN 1854-6250 Journal home: apem-journal.org Original scientific paper Performance modelling based on value analysis for improving product development process architecture Yin, F.P.ab, Gao, Q.a b *, Ji, X.ab aSchool of Mechanical Engineering, Shandong University, Jinan, China bKey Laboratory of High Efficiency and Clean Mechanical Manufacture, Shandong University, Ministry of Education, Jinan, China A B S T R A C T A R T I C L E I N F O Improving the architecture of product development process (PDP) is an effective approach to improve PDP performance. However, performance is difficult to model because the criterion of performance such as development cost, time and product quality are usually contradictory. The objective of this paper is to use process value as the evaluation indicator of PDP performance. The process value of PDP, as well as the ratio of process function and process cost, is discussed and its quantitative method is proposed. The process function is defined as the process effectiveness which considers the importance of each activity of PDP, and its evaluation methods based on rework theory and quality function deployment (QFD) are given. The simulation method is used to illustrate the proposed model and analyze the relation between architecture and process value of PDP, and an optimization model for PDP architecture is provided. With the model, we can get a suitable PDP architecture to balance the cost and product function during product development. © 2017 PEI, University of Maribor. All rights reserved. Keywords: Product development process Process performance Process architecture Value analysis Effectiveness Modelling *Corresponding author: gaoqi@sdu.edu.cn (Gao, Q.) Article history: Received 20 October 2016 Revised 15 February 2017 Accepted 18 February 2017 1. Introduction Product development process (PDP) refers to the entire set of activities to convert customers' needs into a technical and commercial solution [1, 2]. The activities and their relationships are described by the process architecture (or network). Analyzing and optimizing the architecture is helpful to build a PDP with high performance, which implies low development cost, short lead time and high product quality [3, 4]. However, these criterions are usually contradictory. For example, higher product quality may mean that the development time is longer or cost is higher. In addition, high performance is an elusive notion [5]. There have been relevant studies on performance measurement in PDP. Huang and Fu [6] proposed a quantitative model based on signal flow graphs (SFG), assessing process from the aspects of time, cost, quality and robustness. Pun et al. [7] proposed a self-assessment model of new product development performance (NPD) using analytical hierarchy process (AHP). Syamil et al. [8] researched the relationships between product development process and process performance, and provided a model of process performance at project level. The cost estimating [9], development quality [10], product performance [11], feasibility [12] and uncertainty [13] etc. have also been studied by some other researchers, which can reflect or affect the performance of PDP. So many measures and evaluation methods for PDP performance lead to an ambiguous 17 Yin, Gao, Ji definition of performance in above literatures. In addition most works focused on performance at project or program level. O'Donnell and Duffy [14] considered that the overall performance in design is determined both by performance of the design solution (related to the product design parameters or product quality), and the performance of the process (related to the duration or cost). In their model, efficiency and effectiveness are as fundamental elements of performance. Efficiency is seen as the relationship between the knowledge that an activity gained and the used resources; while effectiveness is related to the degree how the outputs of knowledge meet the goals of activities. The model provides an improved perception of performance measurement, but the relationship between efficiency and effectiveness is not direct, because their focuses (efficiency focuses on process, and effectiveness focuses on solution or product) are different. Accordingly in some cases it is difficult to achieve combination of the two variables. With successful application and good effects in manufacturing process, lean thinking has been gradually extended to development field in enterprises. Many works have been reported recently [15-17]. Value is the core concept in lean product development (LPD). The essential of value is to guide improvement processes [18]. Some scholars have tried to improve the performance of PDP through the value analysis method [19] or value stream method [20, 21]. Browning [22] pointed out that value is a function of both the product recipe and the process that produces it. That is to say both the effectiveness of the product and the efficiency of the process affect value. However, there are two types of value in product development, product value and process value according to Chase [23]. Unfortunately, most of researches in this field have ignored the difference between them, and mainly focused on product value. Different from the product value which is defined as a capability provided to a customer at the right time at an appropriate price, process value is defined as the ability to perform with maximum quality at minimum cost. The quality is related to effectiveness, while the cost is related to efficiency, and waste in PDP can be understood as inefficiency and ineffectiveness [18], so the process value provides a new way to evaluate and improve the performance of PDP. This paper is concerned with performance modelling by building the evaluation method of process value. The proposed model that describes the process value of PDP integrates the performance parameters, such as development time, process cost and product quality or function. It involves the factors such as process structure, rework constraints, customer requirements and so on. In the model, we extend O'Donnell and Duffy's [14] concept of effectiveness. In our work, the process effectiveness of one activity is evaluated by its rework. The rework evaluation method based on Design Structure Matrix (DSM) and the creation goal evaluation method based on Quality Function Deployment (QFD) are introduced to the model. Through the optimization of the process value, PDP can be improved from different aspects, such as resource allocation, process architecture improvement, requirement optimization, etc. In this study we focus on improving PDP from the aspect of process architecture improvement. The remainder of this paper is organized as follows. Section 2 discusses the process value and its quantitative method. In section 3 the relationship between architecture and performance of PDP is verified by simulation based on a sample, and a new optimization model for PDP architecture is proposed. An example and some discussions are given in section 4. Finally, the conclusions and some extended research are presented in section 5. 2. Performance modelling based on value analysis 2.1 Process value The importance of one activity in PDP describes the role that the activity plays during developing product, which can be considered as the process creation goal of the activity divided from the overall process creation goal of PDP. We use g to denote the creation goal of an activity and G to denote the overall process creation goal of PDP. The relationship between them can be expressed as Eq. 1. 18 Advances in Production Engineering & Management 12(1) 2017 Performance modelling based on value analysis for improving product development process architecture n G = YJ9i [1) i=l Here n indicates the number of activities. Whether an activity is able to achieve its goals or play its role in PDP depends on the effectiveness of the activity. In our method, the process effectiveness refers to the degree of the relevance between the creation and the goal of the process or activity. The effectiveness of the activity is denoted by e, while the effectiveness of the PDP is denoted by E. The process function can be considered as the actual creation of activity or PDP. The process function of an activity denoted by f and the process function of PDP denoted by F can be calculated as Eq. 2 and Eq. 3. f = g^e (2) F = G • E (3) In addition, /and F have the following relationship: it -I* í = l (4) In our model the process value is defined as the capability to achieve the process creation goal under appropriate cost. The process value of an activity represented by the letter v, and the process value of the PDP represented by the symbol vproc, are formulated as Eq. 5 and Eq. 6. v = yproc _ 9_e_ c G a1 •C + a2 •T (5) (6) The cost of PDP includes both resource consumption C and time consumption T. In Eq. 6, a1 and a2 respectively represent the weight of the cost and development time. The performance model of PDP with effectiveness and process value is expressed in Fig. 1. F=Output-Input Input Product Development Process (PDP) C5> i inputi C¿>...... jgi /i = output\-inpitt\\ OUpUt\ Activity 1 I :: j inputj Activity i ouput\ Output C+T Fig. 1 Effectiveness and process value Advances in Production Engineering & Management 12(1) 2017 19 Yin, Gao, Ji The first part-reflects the different development modes and development methods. The second part E reflects the uncertainties in a PDP, such as the different architectures, reworks and so on. So we can select suitable product development type and improve the architecture of PDP according to Eq. 6. This paper only studies the improvement of architecture under a fixed product development type. The parameters G and C can be considered as the static information, which can be obtained through prediction. Non-dimensional treatment for them is produced in analysis process. The cost C is the accumulation of the cost of all activities in PDP in normal executing case. It is the expected cost, rather than the actual cost of execution, and can be calculated according to Eq. 7. n c i = 1 It = (7) Here ct is the expected cost of unit time of activity i, and di is the expected working time of activity i. Generally, coefficient of C is 1. The development time T is affected by the development process architecture. Assume that the development time on critical path of a PDP is Tmin, the coefficient of T can be calculated by Eq. 8. T = (8) * expected. Here Texpected is the expected time of the product development project, and can be considered as a constant. In Fig. 1 input and output of activity or PDP are indeterminate and difficult to evaluate, so in the next section we will explore the decomposition of G and evaluate E. 2.2 Decomposing of process creation goal G of PDP is the performance or function of product that must be created. As a broad concept, it reflects the extent of how much the product to meet the customer requirements, including technical parameters, quality and other attributes. So that is to say, it is based on the customer requirements and will not be changed during development. In other words, it can be seen as a constant. However, g of an activity is decomposed from G of PDP based on its importance during the development, and difficult to quantify. The coefficient of activity (COA) is used to express the importance of an activity during development. Customer requirement is the start to evaluate process. On referring to Quality Function Deployment (QFD) method, which considers that meeting customer requirement is the ultimate goal of product development, this paper establishes the mapping matrix from customer requirement to product function, and the mapping matrix from product function to activities. Through these matrices the COA can be evaluated. Because the QFD has been quite mature, it is assumed that the customer requirements and all of the mapping matrices are able to be estimated or set by experienced engineers. The customer requirements can be expressed by CR = [CR1,CR2,...,CRi], where I is the total number of customer requirements, and £fc=i CRk = 1. The mapping matrix from product requirements to product function is r/ll rflm RF = rfn - rflr¡ where rfks indicates the importance coefficient of product function s to realize the customer requirement k. The sum of every row in RF is 1. The weight of product functions W is expressed by [W1,W2,..., Wm]. So the evaluation coefficient of product functions (COF) expressed by vector [COF1,COF2,...,COFm] can be calculated with Eq. 9. 20 Advances in Production Engineering & Management 12(1) 2017 Performance modelling based on value analysis for improving product development process architecture COF = (CR •RF)xW (9) Similarly, the matrix FA = fan -fan i falr fanr with sum of every column being 1, expresses the mapping matrix from product function to development process, and fais indicates the importance coefficient of activity i to realize the product function s. So the evaluation coefficient of product function target of activities (COA) expressed by vector [COA1,COA2,...,COAn] , can be calculated with Eq. 10. COA = COF • FAt (10) Then the decomposing formula of G is given as Eq. 11. gt = Gx COAt (11) 2.3 Effectiveness evaluating based on R-DSM The effectiveness reflects the degree of completing the process creation goal. If the creation goal cannot be fully completed by the PDP or the activity, the part that has not been completed can be considered as ineffective, which should be retrieved by rework process. So we can evaluate the effectiveness of activity through its rework. We suppose that the completion degree of the process function of the activity is related to the cost and time the activity consumed. As shown in Fig. 2, if the consumed time and cost is c and the process creation goal is g, it will get the process function f = g • e. In other words, there will be g — f process function left to be reworked, which is expressed as rg. Similarly, in order to get the process creation goal g — f function which can be the process creation goal of the rework, the consumed time and cost (rc) will be iS-Hl which is the rework effort. So the rework rate of 3 an activity can be calculated as r = 9 9 e = 1 — e. Therefore the activity effectiveness can be represented and calculated simply by the rework rate of activity as Eq. 12. 10 Male 90 64 25 54 233 Female 36 20 14 3 73 Total 126 84 39 57 306 Table 3 shows surveyed industries and their size, measured by number of employees. Note that only 278 of the 306 participants reported such information. In this sense, the table shows that 93 automobile and electronics manufacturers were surveyed, while 51 of the studied maquiladoras belonged to the aeronautics sector. However, the medical or surgical sector was the least prominent industry, with only 21 reported cases. As regards company size, all surveyed maquiladoras can be considered large, since they reported more than 500 employees in operation. Table 3 Industrial sector and size of companies surveyed Industrial sector Number of employees Total 1-50 51-100 101-200 201-500 >501 Automotive 3 3 6 9 72 93 Electronics/Electrical 7 8 8 17 53 93 Plastic 1 0 1 3 10 15 Packaging 1 0 1 1 2 5 Aeronautics 15 8 8 5 15 51 Medical 0 21 0 0 0 21 Total 27 41 24 35 152 278 3.2 Questionnaire validation For data validation, we measured reliability of latent variables using seven indices as described in the methodology section. Results from this data validation process are shown in Table 4. Values obtained for R-squared and adjusted R-squared indices demonstrated that latent variables had enough predictive validity. Likewise, since values of the composite reliability index and the Cronbach's alpha were above 0.7, we concluded that all latent variables had sufficient internal consistency. Also, we found enough discriminant validly and no collinearity problems in data, since AVE values were above 0.5 and VIF values were below the threshold (3.3). Finally, since we obtained Q-squared values similar to their corresponding R-squared values, we demonstrated that all latent variables showed predictive validity from both parametric and non-parametric perspectives. Table 4 Statistical validation of latent variables Index Latent variable ICT Agility Operating performance Economic performance R-squared 0.397 0.627 0.459 Adjusted R-squared 0.395 0.624 0.455 Composite reliability 0.933 0.889 0.916 0.874 Cronbach's alpha 0.916 0.813 0.817 0.819 AVE 0.667 0.729 0.846 0.581 VIF 1.757 2.142 2.691 2.718 Q-squared 0.400 0.627 0.453 34 Advances in Production Engineering & Management 12(1) 2017 The impact of information and communication technologies (ICT) on agility, operating, and economical performance... 3.3 Descriptive analysis of items Table 5 shows results from the descriptive analysis of items, which are sorted in descending order based on their median values. According to participants, item Effective use of the Internet in B2B commerce is the most important ICT activity (median value = 4.26). However, in terms of SC Agility, the most valuable feature is Ability to respond to unexpected customer demands (median value = 4.38). In the case of SC Operational Performance, managers from Mexican maquiladoras considered item In-full and on-time deliveries as the most important activity, with a median value of 4.39. Finally, as for Economic Performance, the sample reported SC costs reduction as the most important element (median value = 4.2). Table 5 Descriptive analysis of items ICT Median IR Effective use of the Internet in B2B commerce. 4.26 1.39 Effective use of the Internet for business management. 4.26 1.32 Displayed information and inter-organizational coordination. 4.18 1.49 Intra-organizational Information Systems for SC coordination and collaboration. 4.18 1.37 Removal of SC intermediaries. 4.18 1.47 Use of the Internet for product customization and collaboration. 4.15 1.45 SC optimization through ECR. 4.10 1.45 Agility Ability to respond to unexpected customer demands. 4.38 1.26 Ability to respond to high market fluctuations. 4.31 1.38 Fast re-engineering process. 4.16 1.44 Operating performance In-full and on-time deliveries. 4.39 1.21 Customer satisfaction (no claims or warnings). 4.26 1.34 High-levelled product customization. 4.18 1.35 Short supplier-customer time cycle. 4.14 1.43 SC visibility. 4.12 1.37 Economic performance SC costs reduction. 4.2 1.39 SC performance contributes to cash flow. 4.12 1.35 3.4 Direct effects and hypotheses validation Figure 2 shows the model evaluated as described in the methodology section. Every effect includes a beta (P) value and a P-value; the former is a measure of dependency, whereas the latter was used to determine statistical significance of effects at a 95 % confidence level. Note that all P-values are below 0.05, demonstrating that all beta parameters were statistically significant and had to remain in the model. The figure also shows the percentage of explained variance in every dependent latent variable (R2). In this sense, we found that Economic Performance was 63 % explained by the three other latent variables, since R2 = 0.63. Similarly, Operating Performance was 46 % explained by SC Agility and ICT, since in this case R2=0.46. As regards model fit and quality indices, we obtained the following results: Average Path Coefficient (APC) = 0.396 (with P < 0.001); Average R-Squared (ARS) = 0.572 (with P < 0.001); Average Adjusted R-Squared (AARS) = 0.571 (with P < 0.001); and Average block VIF (AVIF) = 2.234 (ideally <= 3.3). Since these indices demonstrated model's adequacy, we could formulate accurate conclusions on the sixth hypotheses proposed in the introduction section. Such conclusions are presented in Table 6, where based on p-values, the highest positive direct effect occurred from ICT on Agility, implying that when the former increased its standard deviation by one unit, the latter increased by 0.63 units (P = 0.63). Advances in Production Engineering & Management 12(1) 2017 35 García-Alcaraz, Maldonado-Macías, Alor-Hernández, Sánchez-Ramírez r! = 0.40 Fig. 2 Evaluated model Table 6 Direct effects - tested hypotheses Hypothesis Independent variable Dependent variable ß- and P-values_Conclusion H1 ICT Agility 0.63 (P< 0.01) Accepted H2 ICT Operating performance 0.25 (P< 0.01) Accepted H3 Agility Operating performance 0.49 (P< 0.01) Accepted H4 ICT Economic performance 0.10 (P= 0.02) Accepted H5 Agility Economic performance 0.18 (P< 0.01) Accepted H6 Operating performance Economic performance 0.59 (P< 0.01) Accepted 3.5 Total indirect effects Indirect effects occur through mediator variables using two or more model segments; they are important, since they can explain relationships that initially seemed to be statistically nonsignificant In this research, Table 7 shows the sum of indirect effects between latent variables, including statistical validation given by P-values, and the effect size, ES. ES is similar to R2 of direct effects, and it expresses the percentage of explained variance in dependent latent variables. As can be observed, ICT have strong indirect effects on both Economic Performance and Operating Performance, with values p = 0.447 and p = 0.307, respectively. Table 7 Indirect effects To From ICT Agility Economic performance Operating performance 0.447 (p < 0.001) ES = 0.247 0.307 (p < 0.001) ES = 0.172 0.29 (p < 0.001) ES = 0.172 3.6 Total effects Total effects are the sum of direct and indirect effects. Table 8 introduces total effects found in each relationship between latent variables, which based on the P-values, were all statistically significant. Also, it was found that ICT had high total effects on all the other latent variables, thereby demonstrating the importance of ICT implementation in SC. Also, note that the highest ICT effect occurred on SC Agility (p = 0.63 units), whereas ICT impact on Operating Performance seemed a little lower (p = 0.561). Also, we found that SC Agility had significant total effects on Operating Performance (p = 0.488) and Economic Performance (p = 0.47). Table 8 Total Effects To From ICT Agility Operating performance Agility 0.63 (p < 0.001) ES = 0.397 Economic performance 0.547 (p < 0.001) ES = 0.303 0.47 (p < 0.001) ES = 0.300 0.594 (p < 0.001) ES = 0.457 Operating performance 0.561 (p < 0.001) ES = 0.315 0.488 (p< 0.001) ES = 0.316 36 Advances in Production Engineering & Management 12(1) 2017 The impact of information and communication technologies (ICT) on agility, operating, and economical performance... 4. Conclusions, industrial implications, and future work In this research, we provided quantitative dependency measures to demonstrate that ICT implementation has effects on SC agility and operating and economic performance. These findings look trivial and with common sense, but those dependence values represent the main contribution given in this paper. Results introduced in Fig. 2 thus validated the six hypotheses as statistically significant, and they can thus support decision-making in Mexican maquiladoras on ICT implementation as a profitable strategy. Findings here reported also support ICT implementation as a source of competitiveness, since they allow companies to increase SC Agility and visibility, which both impact on Operating Performance and Economic Performance. In this sense, we found that the ICT effects on Economic Performance only increased when SC Agility and Operating Performance were present, since the indirect effect in this relationship was higher than the direct effect (P = 10 vs. B = 0.447). Similar results were reported in the wine industry by Garcia-Alcaraz et al. [3] and Martínez-Loya et al. [4]. Similarly, this study argues that ICT allow for a faster response to customer needs by streamlining changes resulting from demand uncertainty. This argument is supported by the relationship found between ICT and Agility, which showed the highest direct effect (P = 0.63), and in which the former explained 40 % of the variance of the latter (R2 = 0.40). In this research we found that indirect effect from ICT on Operating Performance given through Agility was higher (P = 0.307) than the direct effect (P = 0.25), and total effects equalled 0.561 units. Similar findings were obtained in the relationship between Agility and Economic Performance, in which the direct effect was 11 units below the indirect effect (P = 0.18 vs. B = 0.29) given through Operating Performance. Such results entail the following industrial implications; • Company executives and SC managers must encourage ICT implementation to meet SC and corporate demands. However, it is equally important to properly plan ICT implementation, and provide adequate training on the use of ICT, since their success impact on SC Agility and Operating Performance. • SC visibility is key to the production processes, since it supports companies in making on-time decisions and rapid production process changes. • Managers must pursuit all operating benefits provided by ICT and SC Agility, since both elements guarantee proper Operating Performance, and thus increase Economic Performance. Finally, as future research, we will seek to provide full explanation for explained variance of latent variables, since R2 obtained in this study did not reach the unit. To achieve this goal, we will analyse technological levels and updates of ICT, as well as the role of support and maintenance equipment in SC performance. Acknowledgements We thank the National Council of Science and Technology (CONACYT) and the Teachers' Professional Development Program (PRODEP) for the financial support granted through projects Thematic Network of Industrial Process Optimization, no. 260320 and Supply Chain Optimization. References [1] Ketikidis, P.H., Koh, S.C.L., Dimitriadis, N., Gunasekaran, A., Kehajova, M. (2008). The use of information systems for logistics and supply chain management in South East Europe: Current status and future direction, Omega, Vol. 36, No. 4, 592-599, doi: 10.1016/j.omega.2006.11.010. [2] Mensah, P., Merkuryev, Y., Longo, F. (2015). Using ICT in Developing a Resilient Supply Chain Strategy, Procedia Computer Science, Vol. 43, 101-108, doi: 10.1016/j.procs.2014.12.014. [3] García-Alcaraz, J.L., Maldonado-Macías, A.A. (2016). Just-in-time elements and benefits, Series Management and Industrial Engineering, Springer International Publishing, New York, USA, doi: 10.1007/978-3-319-25919-2. [4] Martínez-Loya, V., García-Alcaraz, J.L., Díaz-Reza, J.R., Marquez-Gayosso, D.G. (2017). The Impact of ICT on Supply Chain Agility and Human Performance, In: Leal-Jamil, G., Lucas-Soares, A., Magalhaes-Pessoa, C.R. (eds.), Advances in Production Engineering & Management 12(1) 2017 37 García-Alcaraz, Maldonado-Macías, Alor-Hernández, Sánchez-Ramírez Handbook of Research on Information Management for Effective Logistics and Supply Chains, IGI-Global, Hershey, PA, USA, 180-198, doi: 10.4018/978-1-5225-0973-8.ch010. [5] Caridi, M., Moretto, A., Perego, A., Tumino, A. (2014). The benefits of supply chain visibility: A value assessment model, International Journal of Production Economics, Vol. 151, 1-19, doi: 10.1016/j.ijpe.2013.12.025. [6] Zhang, A.N., Goh, M., Meng, F. (2011). Conceptual modelling for supply chain inventory visibility, International Journal of Production Economics, Vol. 133, No. 2, 578-585, doi: 10.1016/j.ijpe.2011.03.003. [7] Yu, M.-C., Goh, M. (2014). A multi-objective approach to supply chain visibility and risk, European Journal of Operational Research, Vol. 233, No. 1, 125-130, doi: 10.1016/j.ejor.2013.08.037. [8] Williams, B.D., Roh, J., Tokar, T., Swink, M. (2013). Leveraging supply chain visibility for responsiveness: The moderating role of internal integration, Journal of Operations Management, Vol. 31, No. 7-8, 543-554, doi: 10.1016/j.jom.2013.09.003. [9] Zhang, H.P. (2015). An agent-based simulation model for supply chain collaborative technological innovation diffusion, International Journal of Simulation Modelling, Vol. 14, No. 2, 313-324, doi: 10.2507/IISIMM14(2)CQ6. [10] Sukati, I., Hamid, A.B., Baharun, R., Yusoff, R.M., Anuar, M .A. (2012). The Effect of Organizational Practices on Supply Chain Agility: An Empirical Investigation on Malaysia Manufacturing Industry, Procedia - Social and Behavioral Sciences, Vol. 40, 274-281, doi: 10.1016/j.sbspro.2012.03.191. [11] Kesen, S.E., Kanchanapiboon, A., Das, S.K. (2010). Evaluating supply chain flexibility with order quantity constraints and lost sales, International Journal of Production Economics, Vol. 126, No. 2, 181-188, doi: 10.1016/j.ijpe.2010.03.006. [12] Gosling, J., Purvis, L., Naim, M.M. (2010). Supply chain flexibility as a determinant of supplier selection, International Journal of Production Economics, Vol. 128, No. 1, 11-21, doi: 10.1016/j.ijpe.2009.08.029. [13] Sabbaghi, N., Sheffi, Y., Tsitsiklis, J.N. (2014). Allocational flexibility in constrained supply chains, International Journal of Production Economics, Vol. 153, 86-94, doi: 10.1016/j.ijpe.2014.01.014. [14] Wallace, S.W., Choi, T.-M. (2011). Flexibility, information structure, options, and market power in robust supply chains, International Journal of Production Economics, Vol. 134, No. 2, 284-288, doi: 10.1016/j.ijpe.2009.11.002. [15] Seebacher, G., Winkler, H. (2015). A capability approach to evaluate supply chain flexibility, International Journal of Production Economics, Vol. 167, 177-186, doi: 10.1016/j.ijpe.2015.05.035. [16] Acar, A.Z., Uzunlar, M.B. (2014). The effects of process development and information technology on time-based supply chain performance, Procedia - Social and Behavioral Sciences, Vol. 150, 744-753, doi: 10.1016/j.sbspro. 2014.09.044. [17] Costantino, N., Dotoli, M., Falagario, M., Fanti, M.P., Mangini, A.M. (2012). A model for supply management of agile manufacturing supply chains, International Journal of Production Economics, Vol. 135, No. 1, 451-457, doi: 10.1016/j.ijpe.2011.08.021. [18] Hudnurkar, M., Jakhar, S., Rathod, U. (2014). Factors affecting collaboration in supply chain: A literature review, Procedia - Social and Behavioral Sciences, Vol. 133, 189-202, doi: 10.1016/j.sbspro.2014.04.184. [19] Kisperska-Moron, D., de Haan, J. (2011). Improving supply chain performance to satisfy final customers: "Leagile" experiences of a polish distributor, International Journal of Production Economics, Vol. 133, No. 1, 127134, doi: 10.1016/j.ijpe.2009.12.013. [20] Rouyendegh, B.R., Baç, U., Erkan, T.E. (2014). Sector selection for ERP implementation to achieve most impact on supply chain performance by using AHP-TQPSIS hybrid method, Tehnicki vjesnik - Technical Gazette, Vol. 21, No. 5, 933-937. [21] Correa-Espinal, A., Gómez-Montoya, R., (2009). Information technologies in supply chain management, DYNA, Vol. 76, No. 157, 37-48. [22] Mensah, P., Merkuryev, Y., Longo, F. (2015). Using ICT in developing a resilient supply chain strategy, Procedia Computer Science, Vol. 43, 101-108, doi: 10.1016/j.procs.2014.12.014. [23] de Treville, S., Shapiro, R.D., Hameri, A.-P. (2004). From supply chain to demand chain: The role of lead time reduction in improving demand chain performance, Journal of Operations Management, Vol. 21, No. 6, 613-627, doi: 10.1016/j.jQm.2003.10.001. [24] Warren Liao, T., Chang, P.C. (2010). Impacts of forecast, inventory policy, and lead time on supply chain inventory - A numerical study, International Journal of Production Economics, Vol. 128, No. 2, 527-537, doi: 10.1016/ j.ijpe.2010.07.002. [25] Li, Y., Xu, X., Ye, F. (2011). Supply chain coordination model with controllable lead time and service level constraint, Computers & Industrial Engineering, Vol. 61, No. 3, 858-864, doi: 10.1016/j.cie.2011.05.019. [26] Leber, M., Weber, C., Adam, F., Leber, M. (2014). Mobile application as an innovative supply chain concept and the impact of social capital, International Journal of Simulation Modelling, Vol. 13, No. 2, 135-146, doi: 10.2507/ IJSIMM13(2)1.255. [27] Pan, J.-N., Nguyen, H.T.N. (2015). Achieving customer satisfaction through product-service systems, European Journal of Operational Research, Vol. 247, No. 1, 179-190, doi: 10.1016/j.ejor.2015.05.018. [28] Dehning, B., Richardson, V.J., Zmud, R.W. (2007). The financial performance effects of IT-based supply chain management systems in manufacturing firms, Journal of Operations Management, Vol. 25, No. 4, 806-824, doi: 10.1016/j.jom.2006.09.001. [29] Elgazzar, S.H., Tipi, N.S., Hubbard, N.J., Leach, D.Z. (2012). Linking supply chain processes' performance to a company's financial strategic objectives, European Journal of Operational Research, Vol. 223, No. 1, 276-289, doi: 10.1016/j.ejor.2012.05.043. 38 Advances in Production Engineering & Management 12(1) 2017 The impact of information and communication technologies (ICT) on agility, operating, and economical performance... [30] Kisperska-Moron, D., Swierczek, A. (2009). The agile capabilities of Polish companies in the supply chain: An empirical study, International Journal of Production Economics, Vol. 118, No. 1, 217-224, doi: 10.1016/j.ijpe. 2008.08.019. [31] Ngai, E.W.T., Chau, D.C.K., Chan, T.L.A. (2011). Information technology, operational, and management competencies for supply chain agility: Findings from case studies, The Journal of Strategic Information Systems, Vol. 20, No. 3, 232-249, doi: 10.1016/jjsis.2010.11.002. [32] Kanellou, A., Spathis, C. (2013). Accounting benefits and satisfaction in an ERP environment, International Journal of Accounting Information Systems, Vol. 14, No. 3, 209-234, doi: 10.1016/i.accinf.2012.12.002. [33] Teittinen, H., Pellinen, J., Järvenpää, M. (2013). ERP in action - Challenges and benefits for management control in SME context, International Journal of Accounting Information Systems, Vol. 14, No. 4, 278-296, doi: 10.1016/j.accinf.2012.03.004. [34] Silvestre, B.S. (2015). Sustainable supply chain management in emerging economies: Environmental turbulence, institutional voids and sustainability trajectories, International Journal of Production Economics, Vol. 167, 156169, doi: 10.1016/j.ijpe.2015.05.025. [35] Gligor, D.M., Esmark, C.L., Holcomb, M.C. (2015). Performance outcomes of supply chain agility: When should you be agile?, Journal of Operations Management, Vol. 33-34, 71-82, doi: 10.1016/j.jom.2014.10.008. [36] Yang, J. (2014). Supply chain agility: Securing performance for Chinese manufacturers, International Journal of Production Economics, Vol. 150, 104-113, doi: 10.1016/j.ijpe.2013.12.018. [37] Yusuf, Y.Y., Gunasekaran, A., Musa, A., Dauda, M., El-Berishy, N.M., Cang, S. (2014). A relational study of supply chain agility, competitiveness and business performance in the oil and gas industry, International Journal of Production Economics, Vol. 147, Part B, 531-543, doi: 10.1016/j.ijpe.2012.10.009. [38] Sanzo, M.J., Vázquez, R. (2011). The influence of customer relationship marketing strategies on supply chain relationships: The moderating effects of environmental uncertainty and competitive rivalry, Journal of Business-to-Business Marketing, Vol. 18, No. 1, 50-82, doi: 10.1080/10517121003717799. [39] Yu, W., Jacobs, M.A., Salisbury, W.D., Enns, H. (2013). The effects of supply chain integration on customer satisfaction and financial performance: An organizational learning perspective, International Journal of Production Economics, Vol. 146, No. 1, 346-358, doi: 10.1016/j.ijpe.2013.07.023. [40] Li, Y., Ye, F., Lin, Q. (2015). Optimal lead time policy for short life cycle products under Conditional Value-at-Risk criterion, Computers & Industrial Engineering, Vol. 88, 354-365, doi: 10.1016/j.cie.2015.07.011. [41] Kaloxylos, A., Wolfert, J., Verwaart, T., Terol, C.M., Brewster, C., Robbemond, R., Sundmaker, H. (2013). The use of future internet technologies in the agriculture and food sectors: Integrating the supply chain, Procedia Technology, Vol. 8, 51-60, doi: 10.1016/j.protcy.2013.11.009. [42] Harris, I., Wang, Y., Wang, H. (2015). ICT in multimodal transport and technological trends: Unleashing potential for the future, International Journal of Production Economics, Vol. 159, 88-103, doi: 10.1016/j.ijpe.2014.09.005. [43] Lin, C., Chow, W.S., Madu, C.N., Kuei, C.-H., Yu, P.P. (2005). A structural equation model of supply chain quality management and organizational performance, International Journal of Production Economics, Vol. 96, No. 3, 355365, doi: 10.1016/j.ijpe.2004.05.009. [44] Chung, K.-J., Liao, J.-J., Ting, P.-S., Lin, S.-D., Srivastava, H.M. (2015). The algorithm for the optimal cycle time and pricing decisions for an integrated inventory system with order-size dependent trade credit in supply chain management, Applied Mathematics and Computation, Vol. 268, 322-333, doi: 10.1016/j.amc.2015.06.039. [45] Yusuf, Y.Y., Gunasekaran, A., Adeleye, E.O., Sivayoganathan, K. (2004). Agile supply chain capabilities: Determinants of competitive objectives, European Journal of Operational Research, Vol. 159, No. 2, 379-392, doi: 10.1016/j.ejor.2003.08.022. [46] Musa, A., Gunasekaran, A., Yusuf, Y. (2014). Supply chain product visibility: Methods, systems and impacts, Expert Systems with Applications, Vol. 41, No. 1, 176-194, doi: 10.1016/j.eswa.2013.07.020. [47] Disney, S.M., Naim, M.M., Potter, A. (2004). Assessing the impact of e-business on supply chain dynamics, International Journal of Production Economics, Vol. 89, No. 2, 109-118, doi: 10.1016/S0925-5273(02)00464-4. [48] Ram, J., Corkindale, D., Wu, M.-L. (2014). ERP adoption and the value creation: Examining the contributions of antecedents, Journal of Engineering and Technology Management, Vol. 33, 113-133, doi: 10.1016/j.jengtecman. 2014.04.001. [49] Hingley, M., Lindgreen, A., Grant, D.B. (2015). Intermediaries in power-laden retail supply chains: An opportunity to improve buyer-supplier relationships and collaboration, Industrial Marketing Management, Vol. 50, 78-84, doi: 10.1016/j.indmarman.2015.05.025. [50] Bevilacqua, M., Ciarapica, F.E., Giacchetta, G. (2009). Business process reengineering of a supply chain and a traceability system: A case study, Journal of Food Engineering, Vol. 93, No. 1, 13-22, doi: 10.1016/j.jfoodeng. 2008.12.020. [51] Borgianni, Y., Cascini, G., Rotini, F. (2015). Business process reengineering driven by customer value: A support for undertaking decisions under uncertainty conditions, Computers in Industry, Vol. 68, 132-147, doi: 10.1016/j.compind.2015.01.001. [52] Blome, C., Schoenherr, T. (2011). Supply chain risk management in financial crises - A multiple case-study approach, International Journal of Production Economics, Vol. 134, No. 1, 43-57, doi: 10.1016/j.ijpe.2011.01.002. [53] Gligor, D.M., Holcomb, M.C., Feizabadi, J. (2016). An exploration of the strategic antecedents of firm supply chain agility: The role of a firm's orientations, International Journal of Production Economics, Vol. 179, 24-34, doi: 10.1016/j.ijpe.2016.05.008. Advances in Production Engineering & Management 12(1) 2017 39 García-Alcaraz, Maldonado-Macías, Alor-Hernández, Sánchez-Ramírez [54] Hair Jr, J.F., Ringle, C.M., Sarstedt, M. (2013). Partial least squares structural equation modeling: Rigorous applications, better results and higher acceptance, Long Range Planning, Vol. 46, No. 1-2, 1-12, doi: 10.1016/j.lrp.2013.01.001. [55] García-Alcaraz, J.L., Maldonado, A.A., Iniesta, A.A., Robles, G.C., Hernández, G.A. (2014). A systematic review/survey for JIT implementation: Mexican maquiladoras as case study, Computers in Industry, Vol. 65, No. 4, 761-773, doi: 10.1016/j.compind.2014.02.013. [56] Avelar-Sosa, L., García-Alcaraz, J.L., Castrellón-Torres, J.P. (2014). The effects of some risk factors in the supply chains performance: A case of study, Journal of Applied Research and Technology, Vol. 12, No. 5, 958-968, doi: 10.1016/S1665-6423(14)70602-9. [57] Kock, N., Verville, J., Danesh-Pajou, A., DeLuca, D. (2009). Communication flow orientation in business process modeling and its effect on redesign success: Results from a field study, Decision Support Systems, Vol. 46, No. 2, 562-575, doi: 10.1016/j.dss.2008.10.002. [58] Tastle, W.J., Wierman, M.J. (2007). Consensus and dissention: A measure of ordinal dispersion, International Journal of Approximate Reasoning, Vol. 45, No. 3, 531-545, doi: 10.1016/j.ijar.2006.06.024. [59] Green Jr, K.W., Inman, R.A., Birou, L.M., Whitten, D. (2014). Total JIT (T-JIT) and its impact on supply chain competency and organizational performance, International Journal of Production Economics, Vol. 147, Part A, 125135, doi: 10.1016/j.ijpe.2013.08.026. [60] Merschmann, U., Thonemann, U.W. (2011). Supply chain flexibility, uncertainty and firm performance: An empirical analysis of German manufacturing firms, International Journal of Production Economics, Vol. 130, No. 1, 4353, doi: 10.1016/j.ijpe.2010.10.013. [61] Yang, S., Albert, R., Carlo, T.A. (2013). Transience and constancy of interactions in a plant-frugivore network, Ecosphere, Vol. 4, No. 12, 1-25, doi: 10.1890/ES13-00222.1. [62] Kock, N. (2013). Using WarpPLS in e-collaboration studies: What if I have only one group and one condition?, International Journal of e-Collaboration, Vol. 9, No. 3, 1-12, doi: 10.4018/jec.2013070101. [63] Ketkar, M., Vaidya, O.S. (2012). Study of emerging issues in supply risk management in India, Procedia - Social and Behavioral Sciences, Vol. 37, 57-66, doi: 10.1016/j.sbspro.2012.03.275. 40 Advances in Production Engineering & Management 12(1) 2017 APEM jowatal Advances in Production Engineering & Management Volume 12 | Number 1 | March 2017 | pp 41-50 https://doi.Org/10.14743/apem2017.1.238 ISSN 1854-6250 Journal home: apem-journal.org Original scientific paper Fuzzy Delphi and hybrid AH-MATEL integration for monitoring of paint utilization Kumar, A.a*, Mussada, E.K.a, Ashif, M.b, Tyagi, D.b, Srivastava, A.K.b aBML Munjal University, Gurgaon, India bMBA-Hero Lead Students, BML Munjal University, Gurgaon, India A B S T R A C T A R T I C L E I N F O This study investigates the unattended aspects of paint utilization selection criteria in industries. In today competitive business environment almost all companies focus towards sustainable manufacturing. The utilization evaluation and selection criteria for paint and its consumption reduction is the top priority for industry. Especially in automotive industries, paint shop stands as a centre for hazardous waste due to wastage of paint and thinner during the painting process. This research work focuses on optimizing consumption of paint by finding most important criteria affecting paint consumption and optimizing the same to achieve maximum paint yield. The study uses the routes of Delphi technique in a fuzzy environment to find out the most important criteria for paint utilization selection, so that maximize utilization and minimize consumption reduction of paint has been achieved. An integrated approach of AHP and DEMATEL methods has been implemented to prioritize the criteria and to familiarize the relationship within criteria. The outcomes of the study substantiate and prove that this study is the best way to select particular paint utilization selection criteria for the paint shop and also to anticipate the optimal level of paint utilization. © 2017 PEI, University of Maribor. All rights reserved. Keywords: Automotive industry Paint shop Optimization Paint consumption and utilization AHP DEMATEL *Corresponding author: anilror@gmail.com (Kumar, A.) Article history: Received 20 July 2016 Revised 13 February 2017 Accepted 15 February 2017 1. Introduction In this modern era, intense competition and globalization are the main approaching criteria for many organizations. Every industry aims to provide best services to the customers by providing after sales service, warranty, repair services besides selling the product. The first and foremost characteristic which attracts the customer is the appearance of a vehicle. This appearance of vehicles comes from the design, finishing, paint, etc. Out of which, painting process plays a centre role in providing not only good appearance to vehicles but also prevents corrosion. That's why all the companies prioritize the painting process in assembly. As the painting process involves in investing more money, it is necessary to understand and investigate the suitability of paints on vehicles. Painting process also requires the use of costly and harmful chemicals and other resources and to enhance the profitability, the consumption of paint has to be reduced. Therefore, the selection of a particular paint supplier needs a lot of scrutinizes, and there are many factors which should be considered before selecting one. This study is an attempt to understand experts' views in the evaluation of paint utilization selection criteria so that they can use their supplier accordingly and how it can help in paint consumption reduction. Many researchers across the globe explored various multi criteria decision making (MCDM) approaches in manufacturing [6], for supplier selection [11], supply chain [10], transportation 41 Kumar, Mussada, Ashif, Tyagi, Srivastava and logistics [13]. However, no attempt has been found in selecting paint utilization criteria. After scrutinizing the available literature and identifying the gaps, objectives have been set. The primary objective of this study work is to investigate the main criteria points for paint utilization criteria consumption in a manufacturing plant, prioritizing criteria and establish an interrelationship in assembly and targeting paint suppliers accordingly. To achieve the objectives of the study, three MCDM techniques have been used. Out of these, the first one is Delphi in a fuzzy environment has been used to capture the ambiguity of the expert opinion during criteria selection; secondly, Analytical Hierarchy process (AHP) has been put into use to assist in quantifying relative priorities for the given set of evaluation criteria. Thirdly for confirmation and as a final check of interrelation among the criteria, DEMATEL method was later put into use. The rest of the study has been organized in different parts. In the first part, introduction to the study has been given followed by the required basic preliminaries. The procedure for finalizing criteria for best paint utilization criteria has been given in the third part. Prioritizing of criteria and construction of a network relationship map has been discussed in the fourth part. Discussion and concluding remarks have been given in discussion and conclusion parts. 2. Preliminaries 2.1 Fuzzy sets The some important definitions of fuzzy sets which we employed in this study are given below: Def. 1. A fuzzy set A is a subset of the universal set X, with mapping ^ -X ^ [0,1], where For the fuzzy set A the function value of ^ (x) is called the 'membership value' of x in A representing the degree of truth that x is an element of the fuzzy set A. Def. 2. the triangular fuzzy number (TFN) of fuzzy set defines as follows. !0, x r, 2.2 Fuzzy Delphi method This section describes the procedure of fuzzy Delphi technique using triangular fuzzy number to capture experts' opinions by using Eq. 1. Wk = (lk,mki,rk) (1) Wk represents the fuzzy number for the criteria k. h, mk, and n can be represented as the minimum, average, and maximum number of experts opinions. The center-of-gravity method is used to calculate the value of Sk by using Eq. 2. Sk = Qk + mk + rk)/3 (2) The principles for final selection of the criteria as follows: (1) If Sk ^ A accept criterion k; (2) If Sk < A omit criterion k. Once the paint selection criteria is selected, the evaluation of each criteria against others criteria is done by experts for this AHP is employed and further to find the interrelationship, DEMETAL is utilized. In the earlier studies, AHP method has been used to find weight of criteria and all criteria are considered independent and is not considered to find cause-effect relationship within the criteria and DEMATEL method has been used for not only capturing the importance but also reveals the cause-effect relation within criteria [3, 7, 19]. A brief description of both the methods is described below. 42 Advances in Production Engineering & Management 12(1) 2017 Fuzzy Delphi and hybrid AH-MATEL integration for monitoring of paint utilization 2.3 Analytical hierarchy process (AHP) method AHP is power full tool for handling multi-criteria factors in decision making, developed by Saaty [20]. If there are n criteria through then nx(n-1)/2 mutually comparisons can do with help of this method. 1-9 point scale is used to obtain expert's preferences about the selected criteria. A pairwise comparisons is formed as a matrix shows in Eq. 3. A = (a„) = «11 a i2 ■ «In" «21 «22 ■ a2n = -anl «21 ■ CLnn. 1 «12 V«12 «22 1/«l n 1/a2 r aln a2n a n-ln 1/an- ln (3) a,j is preference comparison the criterion i with criterion j. Eigenvalues and eigenvectors with Eq. 4, are used to calculate the relative weights of the criteria. = Àmaxw (4) Here eigenvector and largest eigenvalue of matrix A, are represented by w and Amax. With the help of Eq. 5 and Eq. 6, the reliability of the judgments of experts has been checked. Consistency Index (CI) = ^max ^ n — 1 CI Consistency Ratio (CR) = — (5) (6) RI represents Random Index and n criteria. The value of RI against the number of criteria is given in Table 1. If CI < 0.1, it shows the consistency of the pairwise matrix and can proceed to calculate final weight of the criteria, otherwise, matrix has to be revised. Table 1 Random index 10 RI 0.00 0.00 0.58 0.90 1.12 1.24 1.32 1.41 1.45 1.45 n 2.4 Decision making trial and evaluation laboratory (DEMATEL) method For better understanding and examining the dependent criteria for an INRM ((Influential Network Relationship Map), DEMATEL technique is generally put into use [8]. Cause computation and each constituent requires proper utilization of the matrix, the structural model and the related mathematical theories in the DEMATEL and thus complex problems are solved easily [18]. Complex and intertwined problem groups are easily solved by the DEMATEL [17]. DEMATEL is generally applied to get a better view of the specific problem and catalyse to the detection of feasible solutions by which we can find out the interdependence between the elements of a system with the help of a casual diagram. The casual diagram portrays the interdependence between the elements within a system. The causal diagrams show contextual relationships rather than graphs without direction and also the strengths of the influence between the elements. The different mathematical steps for DEMATEL as follows: Step 1: Experts have been asked to rate the relationship among the criteria with the scale of 0-5, 0-no effect and 5-high effect. The average of experts' opinion has been calculated by Eq. 7. H A = hj]=1JJ4 (7) k=1 Advances in Production Engineering & Management 12(1) 2017 43 Kumar, Mussada, Ashif, Tyagi, Srivastava Step 2: The matrix normalization has been achieved by Eq. 8. F = mxA, where, m = min ,i,j e {1,2, ...,n] maxi Y!l= i aij' maxj £™=1 ai} Step 3: Eqs. 10-11 have been utilized to estimate total relation matrix T. T= lim (F1 + F2 + - + Fm)=yf¡ m=l where, ^Fl = F1 + F2 + --- + Fm m=l = F( I + F1 + F2 ...+ F™"1) = F(7 - F)_1(7 - F)(7 + F1 + F2 ...+ Fm_1) = F(/-F)"1(1 -F)m T = F(7 -F)"1 After identifying matrix T, r and c with help of Eq. 12 and Eq. 13 are calculated. r~ [rilnxl - c~ [c¿]lxn - it x< J=1 - n X L¿ = 1 nxl lxn Step 4: Eq. 14 had been used to calculate the threshold value (a) and avoid minor effects. n n i=l7=1 (8) (9) (10) (11) (12) (13) (14) where, N elements in the matrix T. 3. Paint utilization selection criteria The fuzzy Delphi technique creates better criteria solutions [2, 5, 14] and used to finalize for paint utilization selection criteria. This concept has been implemented to measure the importance of the criteria by using linguistic scales in the form of TFN [9, 16] as mentioned in Table 2. Table 2 The linguistic scales Linguistic scales Extremely important Important Normal Unimportant Extremely unimportant TFN 0.7 0.9 0.9 0.5 0.7 0.9 0.3 0.5 0.7 0.1 0.3 0.5 0.1 0.1 0.3 44 Advances in Production Engineering & Management 12(1) 2017 Fuzzy Delphi and hybrid AH-MATEL integration for monitoring of paint utilization Table 3 Best paint utilization selection criteria Criteria S Transfer Efficiency (C1) 0.72201 Solid Content of Paint (C2) 0.67320 Conductivity of Paint (C3) 0.65402 Hiding Power (C4) 0.72041 Technical Support (C5) 0.62430 Paint Workability (C6) 0.69831 Thinner Intake (C7) 0.76343 Supply Viscosity (C8) 0.65340 After using this method, the important criterion is generally shifted from the evaluation result and the shifting threshold value effects the number of criteria. This study adopts a threshold value of 0.6. 12 production managers from paint assembly automotive have been interview about paint utilization selection criteria and after using Eq. 1 and Eq. 2, and principles selection criteria are shown in Table 3. Both qualitative and quantitative criteria are considered in evaluation and selection of the best paint utilization criteria and a brief description of each criterion as follows. 3.1 Transfer efficiency Transfer efficiency of a painting process is the comparison of amount of paint deposited on component to the amount of total paint sprayed through the painting gun. This is commonly described as the % of weight of solids sprayed to the weight of solids increased by the component. As an illustration, 70 % transfer efficiency means that 50 % of the weight of the solids in the material that was sprayed actually touched the component and the remaining 30 % was lost during the spray finishing process. With help of formula below, we can calculate transfer efficiency easily. Actual paint deposited on component Transfer efficiency =----- Total paint sprayed on component Transfer efficiency of a painting system plays major role to optimize paint consumption, as we can't achieve paint consumption beyond the transfer efficiency of the painting system by all means. 3.2 Solid content of paint A conventional paint is a mixture of resins, solvents, pigments and additives. When a paint is applied over a surface of any solid portion, a dry or solid portion is left over when the paint is completely dried. The volume of the paint that is left over is represented in terms of volume solid. The volume solid of a coat is the ratio of the volume of non-volatile components to its total wet volume. The figure is generally articulated as percentage. Awareness of Volume Solids provides many benefits: 1) It helps to compare and understand the true cost of different paints, 2) It helps to determine and predict how much paint is actually required to be applied to obtain adequate coverage, and 3) It helps to control the actual quality of the painting. 3.3 Thinner intake of paint Thinner intake of paint is the thinner volume essential to achieve desired viscosity from supplied viscosity. Paint is transported from manufacturing plants to automotive paint shop in with a viscosity ranging from 0.14-0.21 Pa-S but paint cannot be applied on parts with the supplied viscosity by any paint applicator whether it is manual painting or electrostatic spraying. For ease of paint application, viscosity is required in the range of 0.03-0.05 Pa-S depending on the paint and application technique and to reduce the viscosity of paint thinner is added. Volume of thinner added (L) Thinner intake (%) = ——---:——— Volume of raw paint (L) Advances in Production Engineering & Management 12(1) 2017 45 Kumar, Mussada, Ashif, Tyagi, Srivastava 3.4 Paint conductivity Paint Conductivity is the measure of charge carrying capacity of paint, it plays major role in electrostatic painting application where painting is achieved due to potential difference between paint and substrate. Electrostatic painting is based on Coulomb's Law i.e. oppositely charged particle attracts each other. In electrostatic painting, paint is given negative charge and substrate is earthed through conveyor. Paint conductivity is measured by resistivity of paint, which is opposite of paint conductivity. An optimum value of paint conductivity is required to achieve maximum transfer efficiency of an electrostatic painting system. Paint conductivity = 1/Paint resistivity 3.5 Hiding power of paint Hiding power of paint is actually the capability of a particular coating to hide the surface on which coating is used. The thumping power is directly linked to the method by which the film is actually applied and also the film thickness. The hiding power of the coating is influenced by the pigments in the binder media. A coating with strong hiding power develops the pigment particles scatter the light so strongly that they hardly reach the substrate. Thus, hiding power is selected as an important parameter for selecting a paint. 3.6 Supply viscosity Viscosity is described as the internal resistance of a fluid to flow and may be considered as a measure of fluid friction. In paint manufacturing and application industry, the very first information available about paint is its viscosity. Paint supplier supplies paint in a relatively high viscous condition. 3.7 Technical support In an automotive paint shop, the process of painting is carried out inside a spraying chamber. This spraying chamber is an effective pressurized enclosed environment, which is generally used to paint parts of a vehicle loaded in a well-designed hangers fixed on a moving conveyer. For maintaining efficient working conditions, the spray booths are equipped with air supply units, air exhaust blowers, LNG heating system and continuous waster scrubber. During the painting process a lot of these parameters are generally required to be optimized by the paint shop in charge or the engineer, on the other hand some parameters related to paint are optimized by the technical support staff of the concerned paint supplier during the painting process. 3.8 Paint workability Paint workability is defined as the ability of paint to spread over the surface and provide uniform thin layer of paint after baking. 4. Prioritization and network relationship map (NRM) To get the weight of criteria, AHP is used and data has been synthesized in excel and then analysed. Let C = {Cjl j= 1, 2,..., n} is the decision criteria set. The data of the available pair wise comparison of n criteria can actually be summarized into an (n x n) evaluation matrix named A in which each element aij (i, j = 1, 2, ...,n) is of weights of the criteria. We would have (n x n) matrixes for every expert, then the geometric mean of all the matrixes was taken to form a geometric mean matrix [4]. The geometric mean has been taken as more ratio properties are involved [1]. Gij = n .í=i V i,j (15) By using Eq. 15, matrix shown in Table 4 has been formed by averaging all the corresponding ranking of each pairwise comparison of all experts/respondents. 46 Advances in Production Engineering & Management 12(1) 2017 Fuzzy Delphi and hybrid AH-MATEL integration for monitoring of paint utilization Table 4 Average matrix Cl C2 C3 C4 C5 C6 C7 C8 Cl 1.00 0.33 3.00 5.00 5.00 3.00 3.00 5.00 C2 3.00 1.00 3.00 5.00 3.00 3.00 5.00 5.00 C3 0.33 0.33 1.00 3.00 3.00 3.00 5.00 7.00 C4 0.20 0.20 0.33 1.00 1.00 3.00 3.00 5.00 C5 0.20 0.33 0.33 1.00 1.00 0.33 1.00 3.00 C6 0.33 0.33 0.33 0.33 3.00 1.00 3.00 5.00 C7 0.33 0.20 0.20 0.33 1.00 0.33 1.00 1.00 C8 0.20 0.20 0.14 0.20 0.33 0.20 1.00 1.00 For normalization matrix shown in Table 5, is obtained by dividing column by the sum of the corresponding column. As a process of cross verification sum of each column is checked if it was 1 or not and the same has been confirmed. Table 5 Normalization matrix Cl C2 C3 C4 C5 C6 C7 C8 Cl 0.34 0.11 0.36 0.32 0.29 0.22 0.14 0.16 C2 0.54 0.18 0.36 0.32 0.17 0.22 0.23 0.16 C3 0.11 0.06 0.12 0.19 0.17 0.22 0.23 0.22 C4 0.07 0.04 0.04 0.06 0.06 0.22 0.14 0.16 C5 0.07 0.06 0.04 0.06 0.06 0.02 0.05 0.09 C6 0.11 0.06 0.04 0.02 0.17 0.07 0.14 0.16 C7 0.11 0.04 0.02 0.02 0.06 0.02 0.05 0.03 C8 0.07 0.04 0.02 0.01 0.02 0.01 0.05 0.03 The average of each row has been taken to find the weight, by doing this the individual weightage of each criteria and rank has been derived and is shown in Table 6. The degree of consistency (CI) and consistency ratio (CR) has been calculated by routes given by Saaty [20] with Eq. 5 and Eq. 6. The CR value obtained is less than 0.1 and it substantiates the acceptability of matrix M. Table 6 Weightage Criteria Weightage Percentage (%) Rank Transfer Efficiency (C1) 0.221 22.1 2 Solid Content of Paint (C2) 0.291 29.1 1 Conductivity of Paint (C3) 0.165 16.5 3 Hiding Power (C4) 0.097 09.7 4 Technical Support (C5) 0.059 05.9 6 Paint Workability (C6) 0.096 09.6 5 Thinner Intake (C7) 0.041 04.1 7 Supply Viscosity (C8) 0.030 03.0 8 To check the interdependent among selected criteria, DEMATEL technique has been utilized. Firstly, the average matrix A is constructed by Eq. 7 as displayed in Table 7. The normalized influence matrix is calculated by Eq. 8 and Eq. 11 the total influence matrix T. In last the NRM is constructed by Eq. 12 and Eq. 13 as displayed in Fig.1. According to step 2, by using Eq. 8 and Eq. 9, we got m is 0.053 and the nominalization matrix F as follows Table 7 Average matrix Cl C2 C3 C4 C5 C6 C7 C8 Sum Cl 0.000 3.000 3.833 2.000 1.000 1.000 2.083 0.083 13.000 C2 1.000 0.000 1.750 3.000 1.000 2.000 3.000 2.000 13.750 C3 1.917 2.000 0.000 2.000 1.000 2.667 2.917 2.000 14.500 C4 2.917 3.000 2.000 0.000 2.000 2.833 2.000 2.000 16.750 C5 1.000 1.000 1.000 2.000 0.000 2.000 1.000 2.000 10.000 C6 2.000 2.917 2.917 1.000 1.000 0.000 1.000 2.000 12.833 C7 2.000 3.000 2.083 2.000 2.917 3.000 0.000 3.833 18.833 C8 1.000 2.917 2.000 1.000 1.000 2.000 3.917 0.000 13.833 Sum 11.83 17.83 15.58 13.00 9.917 15.50 15.917 13.917 A Advances in Production Engineering & Management 12(1) 2017 47 Kumar, Mussada, Ashif, Tyagi, Srivastava F = 0.000 0.053 0.102 0.155 0.053 0.106 0.106 0.053 0.159 0.000 0.106 0.159 0.053 0.155 0.159 0.155 0.204 0.093 0.000 0.106 0.053 0.155 0.111 0.106 0.106 0.159 0.106 0.000 0.106 0.053 0.106 0.053 0.053 0.053 0.053 0.106 0.000 0.053 0.155 0.053 0.053 0.106 0.142 0.150 0.106 0.000 0.159 0.106 0.111 0.159 0.155 0.106 0.053 0.053 0.000 0.208 0.004 0.106 0.106 0.106 0.106 0.106 0.204 0.000 According to step 3, by using Eq. 11 matrix T is calculated and given as follows 0.250 0.499 0.493 0.373 0.262 0.374 0.431 0.306- 0.319 0.389 0.421 0.427 0.280 0.441 0.492 0.416 0.367 0.498 0.352 0.393 0.284 0.477 0.499 0.422 T - 0.437 0.581 0.486 0.332 0.349 0.518 0.497 0.450 i — 0.240 0.329 0.293 0.299 0.159 0.339 0.302 0.319 0.331 0.483 0.442 0.314 0.247 0.304 0.375 0.374 0.424 0.626 0.524 0.459 0.417 0.568 0.448 0.572 0.314 0.524 0.432 0.343 0.281 0.441 0.535 0.325 - Bold component of matrix are > a. Using Eq. 6 to Eq. 7, Table 8 is found out. Table 8 Cause and effect Crietria ri ct rí + Cj ri -Cj Impact Transfer Efficiency (Ci) 2.989 2.681 5.671 0.308 Cause Solid Content of Paint (C2) 3.185 3.929 7.114 -0.744 Effect Conductivity of Paint (C3) 3.292 3.442 6.735 -0.149 Effect Hiding Power (C4) 3.651 2.942 6.593 0.709 Cause Technical Support (C5) 2.280 2.280 4.560 -0.001 Effect Paint Workability (C6) 2.871 3.463 6.334 -0.593 Effect Thinner Intake (C7) 4.038 3.580 7.618 0.457 Cause Supply Viscosity (C8) 3.196 3.185 6.380 0.011 Cause A threshold value has been set up to obtain the Network Relation Map (Fig.1). The threshold value (a) has been computed by Eq. 14 and that threshold value is also used to remove some minor effects elements in matrix T. In matrix T the values of t^- have been calculated, if element of matrix T greater than threshold value a (0.398) that element shown in bold in matrix T e.g. the value of t12 (0.499) > a (0.398), the arrow in the digraph is drained from Ci to C2. The network relationship map for all the eight criteria is built as depicted in Fig. 1. a IL IL = = 25.50/64 = 0. 398 í=17=1 Hiding Power (CjJIS.W. (1.7I| Thnhold Valu« Alpha-IIJ98 Solid Conti nt of Point (Cj)|7.11, -0.74) Conductivity orPHjnl 1C3H0 15) Fig.1 Network relation map (NRM) within criteria 48 Advances in Production Engineering & Management 12(1) 2017 Fuzzy Delphi and hybrid AH-MATEL integration for monitoring of paint utilization 5. Discussion It is well known acceptable fact that all the components of paints are required, but all of them don't have the same priority. In many cases, decisions are made upon giving crisp values, but all these crisp values are not such an accurate reflection of what is exactly happening in the real world. The need for fuzzy set theory arose from the fact that human judgments about preferences are always unclear and are hard to estimate from numerical values. To handle such situations which are complex and uncertain Delphi method was first applied in the fuzzy environment to finalize the selection criteria of paint selection. AHP was used to prioritize and from the results derived from AHP. DEMATEL technique has been employed to find out the interrelationship relationship between criteria and to identify the interrelationship within the criteria. AHP analysis shows that the criteria solid content of paint has the first rank with the weightage value of 29.1 followed by transfer efficiency, conductivity of paint with the weightage value of 22.1 and 16.5. Hiding power (9.7), paint workability (9.6), technical support (5.9), thinner intake (4.1), and supply viscosity (0.3) are contained the rank four to eight. The oval prioritization of criteria is as C2 > C1 > C3 > C4 > C6 > C5> C7> C8. On the basis of (r-c) values all eight criteria have been divided into two group, i.e., (i) cause group and (ii) effect group. (i) Those criteria has (r-c) has positive value, there are in net cause group and affect the rest criteria, high value shows major impact. The criteria: Transfer Efficiency (C1), Hiding Power (C4), Thinner Intake (C7), and Supply Viscosity (C8) are in this group and having 0.3088, 0.7098, 0.4573 and 0.0110 values. The analysis shows that Hiding Power (C4) is the most critical criteria on the others followed by Thinner Intake (C7), Transfer Efficiency (Ci), and Supply Viscosity (C8). The analysis shows that mutual interaction between Transfer Efficiency (C1) and Thinner Intake (C7) has also the mutual interaction with t17(0.4313) and t71(2.2757), t17(0.4242) values and all are greater than a (0.3985). (ii) If (r-c) has negative value, say it is net receive and all net receive criteria. The current study, Solid Content of Paint (C2), Conductivity of Paint (C3), Technical Support (C5), and Paint Workability (C6) are categorized in the effect group, with the (r-c) values of -0.7442, -0.1499, -0.0001 and -0.5925. The criteria Thinner Intake (C7) is impacting all other criteria followed by Hiding Power (C4), Supply Viscosity (C8), and Transfer Efficiency (C1). 6. Conclusion Paint suppliers optimization and choosing them is a major challenge for any automobile industry. It is important because they have to sustain themselves in the competitive environment and to do that they are always on a look out to upgrade their selection criteria of paint suppliers so that they can achieve their maximum utilization of paint that they usually consume. It has been observed that solid content of paint is the most important criteria and supply viscosity of Paint is the least important dimension for the paint supplier selection. The outcomes of DEMATEL results validates that the thinner intake is the most influential and has the strongest connection to other criteria. The criteria Transfer Efficiency is at rank two with weight 22.1 %, and comes in cause group, which has a direct effect on the criteria i.e. Solid Content of paint, Conductivity of Paint, and Thinner Intake, this the most suggested criteria according analysis of the study for paint supplier selection. The criteria Hiding Power ranks four with weightage 9.7 % and has a direct relationship with Transfer Efficiency, Solid Content of paint, Conductivity of Paint, Paint Workability, Thinner Intake, and Supply Viscosity. Thinner Intake is also in cause group and directly affected by all the other criteria Transfer Efficiency, Solid Content of paint, Conductivity of Paint, Hiding Power, Technical Support, Paint Workability, Thinner Intake, Supply Viscosity. The last criteria, which comes in cause group is Supply Viscosity and affected to Solid Content of paint, Conductivity of Paint, Hiding Power , Paint Workability, Thinner Intake. Advances in Production Engineering & Management 12(1) 2017 49 Kumar, Mussada, Ashif, Tyagi, Srivastava The current work is having a great future scope and more detailed investigations of criteria's that effects the utilization of paint consumption and its selection criteria. Further artificial intelligence techniques could be employed to optimize the various paint utilization criteria. References [1] Büyükozkan, G., £iffi, G. (2012). A combined fuzzy AHP and fuzzy TOPSIS based strategic analysis of electronic service quality in healthcare industry, Expert Systems with Applications, Vol. 39, No. 3, 2341-2354, doi: 10.1016/j.eswa.2011.08.061. [2] Chang, P.-T., Huang, L.-C., Lin, H.-J. (2000). The fuzzy Delphi method via fuzzy statistics and membership function fitting and an application to the human resources, Fuzzy Sets and Systems, Vol. 112, No. 3, 511-520, doi: 10.1016/S0165-0114(98)00067-0. [3] Chen, H.H., Kang, H.Y., Xing, X., Lee, A.H.I., Tong, Y. (2008). Developing new products with knowledge management methods and process development management in a network, Computers in Industry, Vol. 59, No. 2-3, 242253, doi: 10.1016/j.compind.2007.06.020. [4] Dalalah, D., Hayajneh, M., Batieha, F. (2011). A fuzzy multi-criteria decision making model for supplier selection, Expert Systems with Applications, Vol. 38, No. 7, 8384-8391, doi: 10.1016/j.eswa.2011.01.031. [5] Dalkey, N., Helmer, O. (1963). An experimental application of the Delphi method to the use of experts, Management Science, Vol. 9, No. 3, 458-467, doi: 10.1287/mnsc.9.3.458. [6] Ghorabaee, M.K. (2016). Developing an MCDM method for robot selection with interval type-2 fuzzy sets, Robotics and Computer-Integrated Manufacturing, Vol. 37, 221-232, doi: 10.1016/j.rcim.2015.04.007. [7] Ho, W.-R.J., Tsai, C.-L., Tzeng, G.-H., Fang, S.-K. (2011). Combined DEMATEL technique with a novel MCDM model for exploring portfolio selection based on CAPM, Expert Systems with Applications, Vol. 38, No .1, 16-25, doi: 10.1016/j.eswa.2010.05.058. [8] Hung, Y.-H., Huang, T.-L., Hsieh, J.-C., Tsuei, H.-J., Cheng, C.-C., Tzeng, G.-H. (2012). Online reputation management for improving marketing by using a hybrid MCDM model, Knowledge-Based Systems, Vol. 35, 87-93, doi: 10.1016/j.knosys.2012.03.004. [9] Ishikawa, A., Amagasa, M., Shiga, T., Tomizawa, G., Tatsuta, R., Mieno, H. (1993). The max-min Delphi method and fuzzy Delphi method via fuzzy integration, Fuzzy Sets and Systems, Vol. 55, No. 3, 241-253, doi: 10.1016/0165-0114(93)90251-C. [10] Lin, Y.-H., Tseng, M.-L. (2016). Assessing the competitive priorities within sustainable supply chain management under uncertainty, Journal of Cleaner Production, Vol. 112, Part 3, 2133-2144, doi: 10.1016/j.jclepro.2014. 07.012. [11] Malik, M.M., Abdallah, S., Hussain, M. (2016). Assessing supplier environmental performance: Applying analytical hierarchical process in the United Arab Emirates healthcare chain, Renewable and Sustainable Energy Reviews, Vol. 55, 1313-1321, doi: 10.1016/j.rser.2015.05.004. [13] Tadic, S., Zecevic, S., Krstic, M. (2014). A novel hybrid MCDM model based on fuzzy DEMATEL, fuzzy ANP and fuzzy VIKOR for city logistics concept selection, Expert Systems with Applications, Vol. 41, No. 18, 8112-8128, doi: 10.1016/j.eswa.2014.07.021. [14] van Zolingen, S.J., Klaassen, C.A. (2003). Selection processes in a Delphi study about key qualifications in senior secondary vocational education, Technological Forecasting and Social Change, Vol. 70, No. 4, 317-340, doi: 10.1016/S0040-1625(02)00202-0. [15] 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. [16] Kumar, A., Dash, M.K. (2016). Using DEMATEL to construct influential network relation map of consumer decision-making in e-marketplace, International Journal of Business Information Systems, Vol. 21, No. 1, 48-72, doi: 10.1504/IIBIS.2016.073380. [17] Wu, W.-W. (2008). Choosing knowledge management strategies by using a combined ANP and DEMATEL approach, Expert Systems with Applications, Vol. 35, No. 3, 828-835, doi: 10.1016/j.eswa.2007.07.025. [18] Wu, H.-H., Tsai, Y.-N. (2012). An integrated approach of AHP and DEMATEL methods in evaluating the criteria of auto spare parts industry, International Journal of Systems Science, Vol. 43, No. 11, 2114-2124, doi: 10.1080/00207721.2011.564674. [19] Jafari-Moghadam, S., Zali, M., Sanaeepour, H. (2017). Tourism entrepreneurship policy: A hybrid MCDM model combining DEMATEL and ANP (DANP), Decision Science Letters, Vol. 6, No. 3, 233-250, doi: 10.5267/ j.dsl.2016.12.006. [20] Saaty, T.L. (2008). Decision making with the analytic hierarchy process, International Journal of Services Sciences, Vol. 1, No. 1, 83-98. [21] Kumar, A., Dash, M.K. (2014). Criteria exploration and multi-criteria assessment method (AHP) of multi-generational consumer in electronic commerce, International Journal of Business Excellence, Vol. 7, No. 2, 213236, doi: 10.1504/IJBEX.2014.059549. 50 Advances in Production Engineering & Management 12(1) 2017 APEM jowatal Advances in Production Engineering & Management Volume 12 | Number 1 | March 2017 | pp 51-61 https://doi.Org/10.14743/apem2017.1.239 ISSN 1854-6250 Journal home: apem-journal.org Original scientific paper Reconsidering production coordination: A principal-agent theory-based analysis b,c Gong, D. c, Tang, M.bc*, Liu, S.bc, Li, Q. aSchool of Economics and Management, Tsinghua University, China bSchool of Economics and Management, Beijing Jiaotong University, China international Center for Informatics Research, Beijing Jiaotong University, China A B S T R A C T A R T I C L E I N F O Production coordination is a common phenomenon in supply chains. Unlike the existing literature, we examine the production coordination problem from the perspective of asymmetric information: how a manufacturer (leading firm) coordinates the relationships with its subsidiary firm(s) and, subsequently, how market returns influence the leading firm's expected utilities, agency cost and the subsidiary firm's expected incomes. We develop an incentive contract model with asymmetric information based on principal-agent theory. Comparative analysis and simulations are conducted to test the model. Results show that the leading firm's expected utilities and agency cost and the subsidiary firm's expected incomes are significantly affected by the subsidiary firm's capability, cost coefficient, absolute risk aversion factor and output variance (common factors); sharp differences among the leading firm's expected utilities and agency cost and the subsidiary firm's expected incomes were found due to different market returns. Thus, the proposed approach (incentive contract model) can help leading firms apply incentives to optimize production modes to obtain production coordination while considering common factors; market returns differences are included in the new model, in contrast to previous approaches. © 2017 PEI, University of Maribor. All rights reserved. Keywords: Principal-agent theory Production coordination Market returns Information asymmetry Incentive *Corresponding author: mincong@bjtu.edu.cn (Tang, M.) Article history: Received 17 November 2016 Revised 20 December 2016 Accepted 9 February 2017 1. Introduction Production coordination is related to demand, inventory status, production plan, production time, promotion plan, demand forecast and sharing transportation routes (Lee, 2000) [1]. With the development of the supply chain, production coordination will become increasingly complex, and production activities of the supply chain cannot be achieved until bodies of resources (the principal: leading firm, and agent: subsidiary firm) are motivated. Therefore, the problem of production coordination urgently needs to be solved. An optimizing algorithm and tools have been developed to solve the production coordination problem; for example, Maticevic, et al. (2007) used the theory of ERP (enterprise resource planning) to achieve internal supply chain coordination for production[2]; Gong, et al. (2015) created a mathematical model and performed a simulation for the resource sharing model's impact on supply chain efficiency[3]; Galic, et al. (2016) put forward multiple criteria solver (MCS) optimizations to solve an asphalt supply chain problem, and simulation results justified that the proposed model can eliminate the lack of an original model [4]. Incentive theory is an important way to solve the production coordination problem. With the background of the global manufacturing network, Jiao, You and Kumar (2006) established an 51 Gong, Tang, Liu, Li agent-based (leading firms and subsidiary firms involved in production coordination) and contracted-based (market demand) model, which is a useful method to address the production coordination problem[5]. Sahay (2003) believed that the mechanism design was important for the production coordination, and the sharing of information, risk and revenue was essential [6]. With the help of cooperative game theory, Nagarajan and Sos"ic'(2008) analyzed the integration problem of production coordination and proposed the theory of 'vision', which provided a new mode of production coordination[7]. Brintrup (2008) established an agent-based, target-based and role-based model to select suppliers to reduce transaction time and increase corporate rev-enue[8]. In order to meet customers' needs, Akanle and Zhang (2008) proposed optimizing the configuration for production coordination, so they built an agent-based model to coordinate bidding and obtained the optimum based on a genetic algorithm[9]. Yi, et al. (2016) combined put options and selective returns in a proposed contract model and constructed a two-echelon supply chain to analyze risk coordination in supply chains[10]. By applying bargaining theory, Saha, et al. (2014) found that inventory-level and retail-price-dependent demand can play an important role in supply chain coordination contracts[11]. Xu, et al. (2015) analyzed consumer return behavior's effect on buy-back contracts in order to coordinate a robust supply chain[12]. To some extent, the focus of production coordination is principal-agent. In production coordination, principal-agent theory can be developed to cope with the contract relations problem. Principal-agent theory is derived from rational choice model, in which the principal's initial actions are available for agent(s) as incentives to help them make decisions that the principal prefers. Principal-agent theory focuses on the responsive decisions of the agents to to the principal's goal and on how the responsive decisions can be mediated by their actions. In order to operate an effective production coordination network, the leading firm should conduct a comprehensive analysis of actors' (agents') decisions (Compte and Jehiel, 2008) [13]. For the analysis of principal-agent problems in the production coordination network, we focus on the understanding of different roles and their power positions (Kulp, 2002) [14]. In the view of adaptive contract design, Ho, et al. (2016) considered Bandit algorithms to solve repeated principal-agent problems in crowd-sourcing markets[15]. When there exists asymmetric information between the principal and agent, the agent must try its best to achieve his own maximum benefit. With asymmetric information, a distinct strategy should be determined due to the agent's hidden intentions. Consequently, adverse selection and moral hazard arise (Herwig and Sascha, 2011) [16]. Therefore, there is a need to formulate proper conduct regulations for production coordination in business activities (Keser and Willinger, 2007) [17]. Rubin, et al. (2016) focused on the principal-agent algorithm itself, and the results of a gift-exchange experiment showed that the introduction of shocks can significantly reduce the likelihood that the agent will fulfill the contract[18]. To sum up, existing literatures focus mainly on resource configuration, production management, the trust mechanism and the agent model, and the full qualitative descriptions of the principal-agent problem (Herwig and Sascha, 2011) [16]. A wide spectrum of goals, such as wages and premiums, are used as the utility function of the principal and agents (Mukherji et al., 2007) [19]. However, there is a gap in the literature related to production coordination and market returns' effect, that is, in asymmetric information, how a manufacturer (leading firm) coordinates the relationships with its subsidiary firm(s) and, subsequently, how market returns influence the leading firm's expected utilities and agency cost and the subsidiary firm's expected incomes. In this paper, based on principal-agent theory, the incentive contract model is constructed in the context of asymmetric information. Additionally, a comparative analysis and simulations are carried out for the leading firm's expected utilities, the subsidiary firm's expected incomes and the leading firm's agency cost. Thus, this chapter is organized as follows: we first conduct a comprehensive review that forms the theoretical foundation of this study. Section 2 discusses applications of principal-agent models to production coordination. Section 3 discusses solutions of principal-agent models for production coordination. Section 4 presents the simulations. Section 5 concludes. 52 Advances in Production Engineering & Management 12(1) 2017 Reconsidering production coordination: A principal-agent theory-based analysis 2. Principal-agent model of production coordination 2.1 The factors related to the model The principal-agent theory focuses on the rational choice model in which the agent has several tasks to fulfill for the principal. In order to fulfill the tasks, the agent has free access to several means and tools. However, the principal expects the agent to make decisions that the principal prefers. Inevitably, objective conflicts between the principal and agent often arise, and both of them want their maximum benefit; in particular, opportunistic practices of deceit and fraud are possible. Mechanism design is effective way of tackling the principal-agent problem by looking for the common factors related to both the principal and the agent in the context of asymmetric information (Lal and Srinivasan, 1993) [20]. The factors related to both actors consist of the leading firm's expected utilities, the subsidiary firm's expected incomes and the leading firm's agency cost. In this paper, one leading firm is assumed (the principal), and one subsidiary firm is assumed (the agent). (1) The leading firm's expected utilities The leading firm provides products or services to its departments, the subsidiary firm is responsible for the provision of products or services, and the total utilities that the product or service brings about will be owned by the leading firm. In order to stimulate the subsidiary firm, the leading firm needs to pay for the subsidiary firm, which is called remuneration. Therefore, the leading firm's utilities are obtained by deducting from the total utilities. The leading firm is risk neutral, and its utilities can be made by market sales (considering the service price only) or profits (considering the service cost and service price), but the market profit method is more scientific. (2) The subsidiary firm's expected incomes The subsidiary firm can obtain incomes by gaining market share or other ways; at the same time, it must pay a certain cost, including economic cost and risk cost, so the subsidiary firm is risk averse, and it will not make enough efforts to conduct innovation due to the structure of income; thus, it is not conducive to the leading firm's utilities. (3) The leading firm's agency cost The subsidiary firm possesses private information, and it will avoid making efforts. Therefore, the leading firm needs to monitor the subsidiary firm, thus guaranteeing that the subsidiary firm makes as much effort as possible to increase the utilities of its products or services. However, the monitoring will inevitably increase the leading firm's agency cost, which includes the risk cost and the incentive cost (referring to the different utilities that the subsidiary firm brings about in the context of symmetric information and asymmetric information). In the production coordination process, the leading firm wants to increase its own utilities and decrease its agency cost, and the subsidiary firm tries its best to avoid working but wants to gain market share. The goal of cooperative production is to look for the common factors related to the leading firm's expected utilities, the subsidiary firm's expected incomes and the leading firm's agency cost. Based on common factors, the leading firm creates industry policy to realize cooperative production (Fig. 1). Fig. 1 The common factors Advances in Production Engineering & Management 12(1) 2017 53 Gong, Tang, Liu, Li In Fig. 1, many factors affect the leading firm's expected utilities, the subsidiary firm's expected incomes and the leading firm's agency cost in production coordination. The common factors related to the leading firm's expected utilities, the subsidiary firm's expected incomes and the leading firm's agency cost can be calculated based on a mechanism design. 2.2 The improved model (1) Market returns model 1 (the price p and cost c are not considered) The utilities U (that yield Q) are the market gains when the subsidiary firm provides products or services to the users with a certain level of effort, and U is subjected to the subsidiary firm's ability, market prosperity and market randomness. Therefore, U = Af(e) + B + 0,where A is the subsidiary firm's ability (A > 0); e is the effort, /(e)is the function, and /'(e) >0, which means that marginal yields are positive and that yields are positive with subsidiary firm ability; /"(e) < 0, which means that the rate of yields is decreasing; The constant B is market prosperity; d is market randomness, and d~N(0,a2)(a2is the output variance). Theorem 1: The linear relation of the model is still reasonable when /(e) = e. Proof: If /(e) = erl, where 0 < r1 < 1, equation /(e) = erl is consistent with the restrictions above. Supposing E = erl, so U = AE + 6 + 0. Additionally, we can see the full linear relations in expression erl ^E ^U, so U = Af(_e) +6 + 0 is simplified as U = Ae + B + d when /(e) = e. The subsidiary firm generates sales with a certain level of effort, so the leading firm needs to pay for the subsidiary firm's work s(U) = a + ^U,where a is the fixed income, ft is the user's share gains. Theorem 2: The subsidiary firm and the leading firm will obtain the same market gain share ft when the gain in market share is normally distributed. Proof: The leading firm wants while the subsidiary firm wants (/>/^and we can obtain = (1) And p~N(n,o2), so n, pu* -£/=1/ /(^)=V2^a (2) The optimum market share gain for the subsidiary firm and the leading firm will be fiu*=M + J^^fy*^- Jfff (3) Finally, the subsidiary firm and the leading firm will gain the same market share ^ (a = 0). Therefore, the leading firm's expected utilities are E(U - s(U)) = -a + E(1-/3)U = -a + (1- /3)(Ae + B) (4) The subsidiary firm's direct cost is C(e) when providing products or services, and C(e) >0, C"(e) >0. We set C(e) = mer2, where m> 0,r2 > 1, and E = erl, so C(e) =mEr3 (r3 = r2/r1>1). C(e) = mEr2 is consistent with the restrictions above, so C(e) = mEr3 can be simplified as C(e) = be2/2 when m = b/2,r3 = 2, where b is the cost coefficient. In addition, the subsidiary firm must pay the risk cost when participating in collaborative incentive contracts, and the subsidiary firm's certainty equivalent profits (/) can reflect its actual income. Set the utilities u = —erI (exponential distribution) for the subsidiary firm, where r is the absolute risk aversion factor, and r > 0, z~N(m, n2). The subsidiary firm's expected utilities are 54 Advances in Production Engineering & Management 12(1) 2017 Reconsidering production coordination: A principal-agent theory-based analysis f~aJ 1 (z-m)z rn¿ E(u) = I -e~rz^^e = (5) 2 2 E(u) = u(J), so =—erO , ¡ = m — lH— . 0~N(0,o2), so random variable -e !1 = a + [>(Ae + B + 6) — be2/2 obeys a normal distribution, and the subsidiary firm's expected profits are EIi = a + fi(Ae + B)-be2/2 (6) 2 The variance of the subsidiary firm's profits is DI1 = fi2o2, and in equation 1 = m — m is the mean (£7i), nis the variance(DZ), so the subsidiary firm's certainty equivalent profits are 1 = a + /3(Ae + B) -be2/2-r/32o2/2 (7) Where rfi2o2/2 is the risk cost. s is the lowest profit that the subsidiary firm requires, and the subsidiary firm will not participate in an incentive contract when the equivalent profit is less than s. Therefore, the prerequisite that the subsidiary firm participates in the incentive contract is a + /3(Ae + B) -be2/2-r/32o2/2 >s (8) Combined with our previous research [21], the production coordination model based on principal-agent theory is E(U - s(U)) = max{-a + (1-/3)(Ae + B)} fargmax{a + fi(Ae + B) -be2/2 -r^2a2/2} (9) 5 I a +p(Ae + B)-be2/2-rp2a2/2 >s (2) Market returns model 2 (the price p and cost c are considered) In model 1, the utilities U are calculated by yield Q; in this model, p is the price of products or services,(p — c)Q is the user's returns, and therefore, leading firm utilities are E(U) = E«p- c)Q — s(Q)) = -a + (1-£)(p- c)(Ae + B) (10) The subsidiary firm's certainty equivalent profits are I = a + /3(p- c)(Ae + B) -be2/2-r(p- c)2/32o2/2 (11) Additionally, the production coordination model based on principal-agent theory is E(U - s(U))=max{-a + (1- £)(p- c)(Ae + B)} (arg max{ a + c)(Ae + B) -be2/2 -r(p- c)2f}2o2/2} (12) S. . I a + fi{p- c)(Ae + B) — be2/2 - r(p- c)2/32o2/2 > s 3. The solution for the production coordination model 3.1 Market returns model 1 (the price p and cost c are not considered) (1) Symmetric information The leading firm can monitor the subsidiary firm in the case of symmetric information, and the leading firm will not motivate subsidiary firm. At the same time, the leading firm must pay the least even if the subsidiary firm wants the most. Based on the game relations, the subsidiary firm will gain the least in the end (s). Therefore, the cooperative production coordination model including the leading firm and the subsidiary firm changes as follows: Advances in Production Engineering & Management 12(1) 2017 55 Gong, Tang, Liu, Li E(U - s(U)) = max{-a + (1-ß)(Ae + 5)} s.t. a + ß(Ae + B)-be2/2-rß2a2/2 = s ( 3 The differentiation for e yields {a + ß(Ae + B)-be2/2 - rß2a2/2 - s}' =0 (14) Aß So e^ = —.In the case of symmetric information, market share gains ß cannot change regardless of whether the subsidiary firm works hard or not, and the leading firm will not motivate the subsidiary firm to gain market share, and the subsidiary firm will obtain unchanged revenue in the end. So, e* «e* = A/b, ß is neglected. Then, a\ «a* = s + A2/2b . (2) Asymmetric information In the case of asymmetric information, the subsidiary firm will try its best to increase s when {a, ß) is given. Do differentiation for e, {a + ß(Ae + B)-be2/2 - rß2a2/2 - s}' =0 (15) so e = Aß/b. Then, the leading firm takes actions to increase its expected utilities when observing e, that is, E(U - s(U)) = max{-a + (1-ß)(Ae + B)} ( e = Aß/b The differentiation for ß yields s ,a + ß(Ae + B)-be2/2 - rß2o2/2 = s (16) {-a + (1-p)(Ae + B)Y =0 (17) so, p1 = (A2 -Bb)/2A2 (18) Actually, a rational subsidiary firm will obtain s at most, that is, a +p(Ae + B)-be2/2-rp2o2/2 = s (19) Both the subsidiary firm and the leading firm will determine /3, and market share gain /3 is known by the leading firm and the subsidiary firm. The differentiation for p yields {a + p{Ae + B)-be2/2 - rp2a2/2}' =0 (20) so, p2 = Bb/(rbo2 - A2) (21) And p1=p2 , B = (A2 -2A2p1)/b, therefore, p = 1/(1 + rbo2/A2) (22) a = s + be2/2 + rp2a2/2 - p(Ae +B) (23) e = A0/b (24) And s(U) = s + be2/2 + rp2a2/2. Eventually, the leading firm's expected utilities (a,p are known) are 56 Advances in Production Engineering & Management 12(1) 2017 Reconsidering production coordination: A principal-agent theory-based analysis E(U - s(U))=B-s + A2/2b(1 + rba2/A2) (25) The subsidiary firm's expected incomes (a, fi are known) are E(s(U)) = s + A2/2b(1+rba2/A2) (26) The subsidiary firm's expected profits must be s regardless of whether it works hard or not, and the leading firm should motivate the subsidiary firm through expected incomes instead of expected profits, so this paper will discuss the subsidiary firm's expected incomes. In the case of asymmetric information, the principal cannot observe the agent's effort. Therefore, there are two additional kinds of agency costs compared with symmetric information: one is the risk cost; another is the incentive cost. In the case of symmetric information, the leading firm's risk cost is 0; in the case of asymmetric information, the leading firm's risk cost is rfi2a2 ra2 " (27) '(i+TT) The incentive cost refers to the difference between high expected revenue with symmetric information and low expected revenue with asymmetric information. For this reason, the subsidiary firm works harder in the case of symmetric information than in the case of asymmetric information. Therefore, the leading firm's expected utilities will decrease in the case of asymmetric information. Therefore, the leading firm's incentive cost is br2oA M2 EiL) = (Ae* +B- C(e*))-(Ae + B- C(*))== 2(1 + r^^2/^2) (28) Then, the leading firm's agency cost is rC°K + £W = 2(1 + W) (29) 3.2 Market returns model 2 (the price p and cost c are considered) Similarly, the leading firm's expected utilities are E(„-s(„)) = (p-c)B-s + 26((1'+"r^2) (30) The subsidiary firm's expected incomes are (P - C)2^2 The leading firm's agency cost is (p — c)2ra2 rc = R + £(t)=2TT7ïW^ (32) 4. The simulation In order to clarify the factors' relations, the paper conducts a mathematical simulation of the leading firm's expected utilities E(U — s(U)), the subsidiary firm's expected incomes £'(s(y)) and the leading firm's agency cost TC (A, b, r, B, s, a2 > 0) based on MATLAB. Advances in Production Engineering & Management 12(1) 2017 57 Gong, Tang, Liu, Li 4.1 The simulation of the leading firm's expected utilities E(U — s(U)), the subsidiary firm's ability A and the absolute risk aversion factor r (1) Market returns model 1 (the price p and cost c are not considered) The simulation setting is shown below: cost coefficient b = 0.5, output variance a2 = 20000, market prosperity B = 10000, subsidiary firm lowest profits = 100000.The MATLAB simulation procedure is as follows: Initialization (B, s, the interval for A: i e [1000,11000], the interval for r: j e [0,10000], array computations (input i, j and function), the interval check (if i< Max{i) and j < Max(j), computing continues, otherwise computing ends) The code for the function is as follows (others omitted): ( HI,/) \ Sfii.j) = ((a(l, i) *a(l, i))/(l + i 10000 * *a(1, i) j))) + (x - y) The relations between E(U — s(U)), A and r are shown in Fig. 2. (33) E(U - sW)) Fig. 2 The relations between E(U — s(U)), A and r (1) Fig. 2 indicates that E(U — is positive with A; E(U — s(U))is negative with r. (2) Market returns model 2 (the price p and cost c are considered) The simulation setting is shown below: p = 100, c = 50; the others are the same as above. The code for the function is as follows (others omitted): ( K1,j) \ Sf(i,j) = (2500 * (a(1, i) *a(1, i))/(1 + I 10000 * *a(1, i) 1))) + ((50 *x)-y) (34) The relations between E(U — s(U)), A and r are shown in Fig. 3. In Fig. 3, the results are similar to those in Fig. 2, but the magnitude of relations between E(U — s(U)), A and r is times greater in Fig. 4 compared with Fig. 3, so p and c affect the model most. 58 Advances in Production Engineering & Management 12(1) 2017 Reconsidering production coordination: A principal-agent theory-based analysis 4.2 The simulation of the leading firm's expected utilities E{U — s(U)), cost coefficient b, and market prosperity B (1) Market returns model 1 (the price p and cost c are not considered) The simulation setting is shown below: subsidiary firm ability A = 5000, output variance a2 = 20000, subsidiary firm absolute risk aversion factor r = 500, subsidiary firm lowest profit s = 100000.The code for the function is as follows (others omitted): Sf(i,j) = (25000000/(2 * 6(1,;) *(1 + 0.4*6(1,;))) + (a(1, 0 - 100000) (35) The relations between E(U — s(U)), B and b are shown in Fig. 4. Fig. 4 indicates that E(U — is negative with 6; E(U — is positive with B. (2) Market returns model 2 (the price p and cost c are considered) The simulation setting is shown below: p = 100, c = 50; the others are the same as above. The code for the function is as follows (others omitted): Sf(i,j) = (2500 * 25000000/(2 * 6(1,;') *(1 + 0.4*6(1,;'))) + (50 * a(1,0 - 100000) (36) The relations between E(U — s(U)), B and 6 are shown in Fig. 5. In Fig. 5, the results are similar to those in Fig. 4, but the magnitude of relations between E(U — s(U)), B and 6 is 104 times greater, so p and c affect the model most Additionally, we find that E(U — s(U))is negative with s; E(U — is negative with a2; and p and c affect the model most Based on the analysis above, the relations between factors {A, 6, r, B, s, a2), E(s(U)^ and TC can be calculated as well. 10 15 10 Fig. 4 The relations between E(U — s(U)), B and b (1) 10 • » b Fig. 5 The relations between E(U — s(U)), B and b (2) Advances in Production Engineering & Management 12(1) 2017 59 Gong, Tang, Liu, Li To sum up, the leading firm's expected utilities E(U — s(U}), the subsidiary firm's expected incomes E(s(U)^ and the leading firm's agency costTCare all affected by the subsidiary firm's ability A, the cost coefficient b, the absolute risk aversion factor r and the output ance o2; E(U — s(U)), E(s(U)) and TCdiffer (p — c)2 times due to the market returns model 1 (the price p and cost c are not considered) compared with the market returns model 2 (the price p and cost c are considered). 5. Conclusion Production networks are an effective way to realize production coordination. The leading firm and the subsidiary firm comprise production networks and enhance production networks' core competencies if parties are united. In the production networks, the leading firm has a special market position due to its access to critical resources. Additionally, the leading firm greatly determines the goals followed by the subsidiary firm in production networks. Inevitably, objective conflicts between the principal and agent often arise, as both of them want their maximum benefit, and in particular, opportunistic practices of deceit and fraud are possible. Mechanism design is an effective way to tackle the principal-agent problem by looking for the common factors related to both the principal and the agent in the case of asymmetric information. The paper develops a principal-agent theory for production coordination. The goal of the principal-agent theory is to analyze the problems occurring between the leading firm and the subsidiary firm and to find the common factors of production coordination, develop solutions for these problems, and the needs of incentive contracts to fulfill the leading firm's requirements and the subsidiary firm's profit. The findings of the simulation help to illuminate the potential behavioral uncertainties between the leading firm and the subsidiary firm and to apply the optimal kinds of measures in production coordination. The leading firm's expected utilities, the subsidiary firm's expected incomes and the leading firm's agency cost are all affected by the subsidiary firm's ability A, cost coefficient b, absolute risk aversion factor r and output variance a2. The leading firm's expected utilities, the subsidiary firm's expected incomes and the leading firm's agency cost differ thousands of times due to the different market returns (the price p and cost c are considered or not). The common factors related to both actors are the leading firm's expected utilities, the subsidiary firm's expected incomes and the leading firm's agency cost; in particular, the difference in market returns is included in the new model, in contrast to previous approaches. Thus, the proposed approach (incentive contract model) can help the leading firm apply incentives to optimize production modes to achieve production coordination while taking into account these common factors and market returns. In the model of Pareto loss, the leading firm and the subsidiary firm will make efforts to renegotiate jointly to increase the effort e (e = fiA/b); in Pareto improvement, the utilities for the leading firm and the subsidiary firm will increase. Therefore, we should improve the subsidiary firm's ability, reduce the subsidiary firm's cost coefficient, lower the subsidiary firm's absolute risk aversion, and pay attention to random market factors. The leading firm can take on incentive measures to optimize production coordination based on the common factors. Acknowledgement The study is supported by a project funded by the China Postdoctoral Science Foundation (2016M591194) and the National Natural Science Foundation (71132008,71390334). We appreciate their support very much. References [1] Lee, H.L. (2000). Creating value through supply chain integration, Supply Chain Management Review, 30-36. [2] Maticevic, G., Lovric, T., Cicak, M. (2007), Using ERP system to improve internal supply chain coordination, Tehnicki vjesnik - Techical Gazette, Vol. 14, No. 3-4, 11-21. [3] Gong, D., Liu, S., Lu, X. (2015). Modelling the impacts of resource sharing on supply chain efficiency, International Journal of Simulation Modelling, Vol. 14, No. 4, 744-755, doi: 10.2507/HSIMM14(4)CQ20. 60 Advances in Production Engineering & Management 12(1) 2017 Reconsidering production coordination: A principal-agent theory-based analysis [4] Galic, M., Zavrski, I., Dolacek-Alduk, Z. (2016), Methodology and algorithm for asphalt supply chain optimization, Tehnicki vjesnik - Techical Gazette, Vol. 23, No. 4, 1193-1199, doi: 10.17559/TV-20150623140015. [5] Jiao, J.(R.), You, X., Kumar, A. (2006). An agent-based framework for collaborative negotiation in the global manufacturing supply chain network, Robotics and Computer-Integrated Manufacturing, Vol. 22, No. 3, 239-255, doi: 10.1016/j.reim.2005.04.003. [6] Sahay, B.S. (2003). Supply chain collaboration: The key to value creation, Work Study, Vol. 52, No. 2, 76-83, doi: 10.1108/00438020310462872. [7] Nagarajan, M., Sosic, G. (2008). Game-theoretic analysis of cooperation among supply chain agents: Review and extensions, European Journal of Operational Research, Vol. 187, No. 3, 719-745, doi: 10.1016/j.ejor.2006.05.045. [8] Brintrup, A. (2010). Behaviour adaptation in the multi-agent, multi-objective and multi-role supply chain, Computers in Industry, Vol. 61, No. 7, 636-645, doi: 10.1016/j.compind.2010.03.010. [9] Akanle, O.M., Zhang, D.Z. (2008). Agent-based model for optimising supply-chain configurations, International Journal of Production Economics, Vol. 115, No. 2, 444-460, doi: 10.1016/i.iipe.2008.02.019. [10] Yi, H.Y., Guo, P. (2015). Modelling risk coordination of supply chains with put option contracts and selective return policies, International Journal of Simulation Modelling, Vol. 14, No. 3, 551-562, doi: 10.2507/IISIMM14 (3)CO15. [11] Saha, S., Goyal, S.K. (2015). Supply chain coordination contracts with inventory level and retail price dependent demand, International Journal of Production Economics, Vol. 161, 140-152, doi: 10.1016/j.ijpe.2014.12.025. [12] Xu, L., Li, Y., Govindan, K., Xu, X. (2015). Consumer returns policies with endogenous deadline and supply chain coordination, European Journal of Operational Research, Vol. 242, No. 1, 88-99, doi: 10.1016/j.ejor.2014.09.049. [13] Compte, O., Jehiel, P. (2008). Gathering information before signing a contract: A screening perspective, International Journal of Industrial Organization, Vol. 26, No. 1, 206-212, doi: 10.1016/j.ijindorg.2006.11.002. [14] Cohen Kulp, S. (2002). The effect of information precision and information reliability in manufacturer-retailer relationships, The Accounting Review, Vol. 77, No. 3, 653-677, doi: 10.2308/accr.2002.77.3.653. [15] Ho, C.-J., Slivkins, A., Wortman Vaughan, J. (2016). Adaptive contract design for crowdsourcing markets: Bandit algorithms for repeated principal-agent problems, Journal of Artificial Intelligence Research, Vol. 55, 317-359, doi:10.1613/jair.4940. [16] Winkler, H., Kraus, S. (2011). Management of principal-agent problems in supply chains, International Journal of Business Research, Vol. 11, No. 2, 12-22. [17] Keser, C., Willinger, M. (2007). Theories of behavior in principal-agent relationships with hidden action, European Economic Review, Vol. 51, No. 6, 1514-1533, doi: 10.1016/j.euroecorev.2006.10.007. [18] Rubin, J., Sheremeta, R. (2015). Principal-agent settings with random shocks, Management Science, Vol. 62, No. 4, 985-999, doi: 10.1287/mnsc.2015.2177. [19] Mukherji, A., Wright, P., Mukherji, J. (2007). Cohesiveness and goals in agency networks: Explaining conflict and cooperation, The Journal of Socio-Economics, Vol. 36, No. 6, 949-964, doi: 10.1016/j.socec.2007.01.024. [20] Lal, R., Srinivasan, V. (1993). Compensation plans for single- and multi-product salesforces: An application of the Holmstrom-Milgrom model, Management Science, Vol. 39, No. 7, 777-793, doi: 10.1287/mnsc.39.7.777. [21] Xu, W.Y., Zhang, Z.J., Gong, D.Q. (2014). Cooperative supply chain management under asymmetric information, Journal of Applied Research and Technology, Vol. 12, No. 2, 182-191 doi: 10.1016/S1665-6423(14)72334-X. Advances in Production Engineering & Management 12(1) 2017 61 Advances in Production Engineering & Management Volume 12 | Number 1 | March 2017 | pp 62-74 https://doi.Org/10.14743/apem2017.1.240 ISSN 1854-6250 Journal home: apem-journal.org Original scientific paper A genetic regulatory network-based sequencing method for mixed-model assembly lines Lv, Y.a, Zhang, J.a*, Qin, W.a aSchool of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai, China A B S T R A C T A R T I C L E I N F O Mixed-model sequencing to minimize work overload at stations is regarded as one of the most concerned optimization problems in assembly lines manufacturing a variety of product models simultaneously. A novel sequencing method based on the genetic regulatory network is proposed to solve this problem. First, genes, gene regulation equations and gene expression procedures are developed in the network based on its similarity with the mixed-model sequencing problem. Each two-state gene represents a binary decision variable of the mathematical model. The gene regulation equations describe decision variable interactions in the constraints and objectives. The gene expression procedure depends on the regulation equations to generate solutions, in which the value of each decision variable is indicated by the expression state of the related gene. Second, regulatory parameter optimization in the regulation equations minimizes the work overload at stations. The effectiveness of the proposed method is validated through experiments consisting of reference instances and industrial instances. The experimental results demonstrate that this method outperforms other methods in large-scale instances. © 2017 PEI, University of Maribor. All rights reserved. Keywords: Assembly line Mixed-model sequencing Work overload Genetic regulatory network Genetic algorithm *Corresponding author: zhangjie@sjtu.edu.cn (Zhang, J.) Article history: Received 11 October 2016 Revised 16 January 2017 Accepted 17 January 2017 1. Introduction Mixed-model assembly lines (MMALs) are widely applied to the industrial engineering world because they can assemble various models of products in a facultative sequence while reducing setup times [1]. Although it is possible to implement any model sequence, the model sequence causes different economic impacts in the actual environment [2]. For instance, different models require diverging processing times at stations to complete the specific assembly operations that realize customized function requirements [3]. The cycle time defines the standard time to process a product at a station, which is typically the average of the processing times of different models weighted by the model demands [4]. The processing times required in complex operations are thus greater than the cycle time, while those required in simple operations are less than the cycle time [5, 6]. If some complex operations are processed continuously by using a specific station, then assembly tasks may not be completed before the operators have reached the down-stream station border, which is regarded as a work overload situation. Although utility workers or line stoppages are adopted to deal with the work overload situation, these lead to additional costs [7]. The mixed-model sequencing (MMS) problem is thus addressed to minimize the work overload at stations, in which different models of products are arranged by performing alternately complex operations and simple ones at each station [8]. 62 A genetic regulatory network-based sequencing method for mixed-model assembly lines The MMS problem has attracted a lot of attention because of its complexity and practical value. Various types of MMS methods including exact solution procedures, heuristic procedures and meta-heuristics have been proposed by scholars. However, the existing methods can hardly achieve high-quality solution when the problem is from large-scale instances and requires acceptable computational time. This study aims to fill in this gap by proposing a novel method based on the genetic regulatory network (GRN). In a GRN, gene states are the same as decision variable values in the MMS problem. Gene regulations describe the interconnection between genes, which have an analogous function with constraints. The gene expression procedure governed by gene regulations determines gene states iteratively, which is similar to a heuristic sequencing procedure. Based on these similarities, genes are first defined in the GRN to represent decision variables. Second, gene regulation equations are developed to express not only hard constraints in the mathematical model, but also soft constraints derived from certain sequencing rules. The importance of soft constraints is weighted by regulatory parameters in the equations. Third, the gene expression procedure is designed to indicate a heuristic procedure that is specified by regulatory parameters. Finally, the regulatory parameters are optimized to obtain the optimal solutions with minimum work overload at stations. Thereupon, the key contribution is the extension of GRN applications to assembly line scheduling. A series of computational experiments are conducted to validate the effectiveness of this GRN-based method. The remainder of this paper is organized as follows. A relevant literature review of MMS methods and GRNs is offered in Section 2. A mathematical model of MMS problems is presented in Section 3. A GRN-based sequencing method is given in Section 4. Section 5 contains the experimental results and discussions. Conclusions and future research directions are discussed in Section 6. 2. Literature review 2.1 Mixed-model sequencing problem In terms of sequencing in MMALs, Boysen et al. [9] provided an integrated review to discuss three fundamental approaches, i.e., MMS, car sequencing and level scheduling. Of these approaches, the MMS methods aim at minimizing sequence-dependent work overload based on a detailed scheduling in MMALs, which have been widely investigated by researchers [10]. These MMS methods can be further classified into four major classes: branch-and-bound computation, exact solution procedure, heuristic procedure, and meta-heuristic [11-13]. The focus of the existing literature was on heuristic procedure and meta-heuristic because it is impractical to implement other methods for large-scale instances [14-16]. For instance, Cano et al. [17] used a scatter search method that selects from 20 priority rules to generate hyper heuristic procedures. Gujjula et al. [18] proposed a heuristic method based on Vogel's approximation method to address large-scale MMS problems. Cortez and Costa [19] developed a set of fast constructive heuristics, two local search procedures and a meta-heuristic to deal with a specific problem featured with worker-dependent processing times. Well-known meta-heuristics such as genetic algorithm and ant colony algorithm have also been employed to solve a variety of MMS problems [20-25]. In general, most heuristic methods used greedy priority rules to construct rapidly a model sequence by appending iteratively alternative models into it. These methods chose products with minimum objective function values at each iteration, which led to intensive increases of the work overload in the last part of the model sequence. Hence, these approaches based on heuristic procedures could obtain good solutions rather than optimal ones. Although other heuristic methods were considered to improve the solution quality (e.g., the work overload increases caused by remaining model copies were taken into account at each iteration), they resulted in the computational effort to be dramatically increased for large-scale problems. Alternatively, meta-heuristic methods could find optimal solutions or near-optimal solutions by globally searching among all the feasible ones. However, the global search caused the computational effort to be increased with the number of model copies as well as the number of stations. Advances in Production Engineering & Management 12(1) 2017 63 Lv, Zhang, Qin Consequently, it is difficult to generate a near-optimal model sequence with an acceptable computational effort for real large-scale instances in the actual manufacturing environment. In this paper, we propose a novel GRN-based method to solve MMS problems. The critical factors of this method include the description of the MMS problem by using a GRN and the integration of certain validated sequencing rules in the GRN. Based on such a GRN, the proposed method can solve the MMS problem more efficiently than meta-heuristics without compromising the solution quality, especially for large-scale instances. 2.2 Genetic regulatory network The GRN is a structured network that describes the regulation of gene expression in cells [26]. It has been widely applied by biologists to investigate the dynamic changes of cell morphologies, and has become a hot topic in the past few years. A GRN has at least the following three elements in common: genes, gene regulations and gene expression procedure [27]. Each gene has two alternative states (i.e., the expressed state and the unexpressed state). If a gene is in the expressed state, it has regulatory effects on the states of other genes, which is the primary form of gene regulations. Based upon these regulations, gene expression procedure iteratively converts certain genes in the unexpressed state into ones in the expressed state if there are enough positive regulatory effects on these genes. Since a cell is mainly composed of copied components (e.g., mRNAs and proteins), gene expression procedure finally determines the cell's morphology in accordance with genes in the expressed states [28]. Various formalisms have already been employed to describe GRNs, for instance, Bayesian networks, directed graphs, partial differential equations, Boolean networks, qualitative differential equations, stochastic equations, and rule-based formalisms [29]. 3. Problem description The following assumptions are taken into consideration when constructing the mathematical model: • The assembly line is a 'moving line' in which the conveyor moves at a constant speed; • The length of a station is a fixed one (measured by the product passing time), and neighboring stations do not overlap; • Products are equi-spaced on the line by launching each other after a constant time interval, which is equivalent to the cycle time; • The operation processing time at a station is not longer than the length of this station; • The impact of unfinished works on operations at succeeding stations is not taken into consideration; • Operators return with infinite velocity to the subsequent product; • The model changeover time is included in the operation processing time; • To facilitate the presentation, the notations listed in Table 1 are used in the development of the mathematical model. The mathematical model takes the following form. T K Minimize t= i fc=i T Wtk (1) (2) (3) s,k = 0 Vk (4) stk >0,wtk >0 Vt,k (5) 64 Advances in Production Engineering & Management 12(1) 2017 A genetic regulatory network-based sequencing method for mixed-model assembly lines M wtk >stk+^ xtmpmk -lk W. k [6] m=l M St+l,k >stk+^ xtmpmk -wtk-c Vt, k [7] m=l Table 1 Problem's notations Notations Definitions Sets {1. Set of products in the model sequence {!.- ...fc,... ,K} Set of stations {1... ,,m,.. ,,M} Set of models Parameters dm c T lk Pmk Demand for model m in the production plan Cycle time Total demand for products, T = £m=i dm Length of station k (time unit) Operation processing time of model m at station fc Variables stk wtk xt.m. Starting time for assembling the tth product at station fc Extra operation processing time for the t-th product at station fc (work overload time) Binary variable: 1, if the t-th product belongs to model m; 0, otherwise The objective of the mathematical model is to minimize the total work overload at stations. The constraints in Eq. 2 ensure that the model sequence satisfies the demand for each model. Eq. 3 makes sure that a station cannot process more than one product at the same time within each production cycle. Eq. 4 represents the initial state of stations. Eq. 5 and Eq. 6 ensure that all the operations should be processed within boundaries of their related stations. If operations on a product cannot be finished within boundaries of their related stations, then a work overload occurs. Eq. 7 makes sure that operators at station k can start processing product (t + 1) after they have completed the operations on product t or product t has left station k. 4. Genetic regulatory network-based sequencing method 4.1 Mapping between the mathematical model and the genetic regulatory network In the mathematical model presented in Section 3, the binary decision variables, the constraints and the solution are similar to the genes, the gene regulations and the gene expression procedure in a GRN, respectively. Based on these similarities, the model can be represented by using a GRN (shown in Fig. 1): (1) each gene represents a decision variable; (2) gene regulations describe constraints; (3) gene expression procedure generates solutions. MMS Problem GRN gene gene expression procedure (gene state: unexpressed^expressed) Fig. 1 Mapping between the MMS problem and the GRN Advances in Production Engineering & Management 12(1) 2017 65 Lv, Zhang, Qin In general, the differential equation method is the most suitable method to develop gene regulation equations because it can represent gene regulations in a quantitative form. In the GRN, first, regulation equations are used to describe all the constraints, and the gene expression procedure is developed based upon the regulation equations to obtain feasible solutions. Second, some soft constraints related to sequencing rules are integrated in the regulation equations to decrease work overload at stations. Third, regulatory parameters are optimized to integrate reasonable sequencing rules and thus minimize work overload at stations. Consequently, as shown in Fig. 2, the GRN-based method contains two parts: • A GRN is developed based on the mathematical model and certain sequencing rules; • Regulatory parameters are optimized by using a genetic algorithm in order to minimize the work overload. Step 2 Regulatory Parameter Optimization optimization algorithm regulatory parameters gene expression procedure " * objective function evaluation J Fig. 2 Outline of the GRN-based method 4.2 Genetic regulatory network establishment According to decision variables of the mathematical model, genes {9tm | t = 1,2, ...,T,m = 1,2,... ,M} are generated. Each gene 9tm indicates that a product of model m is assigned to the tth position of the model sequence. Moreover, the regulation equation in Eq. 8 is developed to express constraints. Vi = fi(Xi(t),X2(t),'~, xI(t),£1,£2,-, eE) (8) where I represents the number of genes in a GRN, xt(t) Vi E {1,2, •••,/} is a binary variable that is equal to 1 if gene is in the expressed state at time t, otherwise, it is equal to 0, is a nonlinear function, vt represents the inhibition coefficient to convert gene 9i to the expressed state, £1,£2,-',£e are regulatory parameters. In terms of Eq. 2 and Eq. 3, the regulation equation first describes following constraints of each position in the model sequence: (1) A model can be selected when other models have not been selected yet. (2) A model can be selected when the demand for this model has not been satisfied at former positions. Moreover, soft constraints related to the study of Cano et al. [17] and the study of Dormer et al. [30] are also included in the regulation equation: (3) A model can be selected if it causes the least work overload at stations. (4) A model can be selected if it leads to the least idle time at stations. (5) A model can be selected if its production ratio best matches its demand ratio in the production plan. No sequencing procedure could satisfy all the soft constraints completely, and each unsatisfied case might increase the work overload at stations. The regulation equation of gene 9tm is thereby developed as follows: K K Vtrn = ¿1 + Pmk ~lk) + Stk ~Vmk) k=l k=l 66 Advances in Production Engineering & Management 12(1) 2017 A genetic regulatory network-based sequencing method for mixed-model assembly lines where vtm represents the inhibition coefficient to gene dtm, H(x) is a step function satisfying H(x) = 0 (x< 0) and H(x) = +œ (x> 0), 0 (x) is a piecewise function satisfying 0 (x) = 0 (x < 0) and 0 (x) = x (x> 0), £x, £2, £3, £4 represent regulatory parameters that combine regulation segments derived from different constraints. The first three terms of the right side of Eq. 9 indicate the inhibition to gene dtm owing to soft constraints (3) to (5), respectively. The last two terms of the right side of Eq. 9 describe constraints (1) and (2), respectively. Constraints in Eq. 4 to Eq. 7 are embodied in the calculation of stk: M 0 if Stk + ^ xtmPmk ^C m=l M M stk + ^ xtmPmk ~c if cT, the model sequence is obtained based on the gene states {xtm |t = 1,2, ...,T, m = 1,2.....M}. Table 2 Pseudo codes of gene expression procedure //initialization for t ^ 1 to T do for m ^ 1 to M do Xtm ^ 0 //all the genes are initialized in the unexpressed state next; next; //gene expression circulation for k ^ 1 to K do slk^0 // initialization of stations next; for t ^ 1 to T do //discrete time m0 ^1, v0 ^ +ot //index of the gene with minimum vtm for m ^ 1 to M do calculate vtm in Eq. 9 //calculate inhibition coefficients if vtm Evaluation: lígnt=m, thenxtm=l, elsexim=0 for t=l,2,—,T —► Eq. 1 —► fitness,, PMuation —► g'«P=g«q g'nq=gnp p,q-rand[l,7] i>=rand[l, M] If = b then g'„t=g„t for i= 1,2,—,7 Crpsspyer: PCrossover —► c=rand[l, A/] t'=l If gnt^ftand^rt ^fithen g'„t=gc(t'++) for t=l,2,---,T Fig. 4 Basic steps to solve the MMS problem by using an ICGA Table 5 Processing times in industrial instances m=1 m=2 m=3 m=4 m=5 m=6 m=7 m=8 m=9 lk k=1 104 100 97 92 100 94 103 100 101 195 k=2 103 103 105 107 101 108 106 102 110 195 k=3 165 156 164 161 148 156 154 164 155 195 k=4 166 175 172 167 168 167 168 156 173 195 k=5 111 114 114 115 117 117 115 111 111 195 k=6 126 121 122 124 127 130 120 121 134 195 k=7 97 96 96 93 96 89 94 101 92 195 k=8 100 97 95 106 94 102 103 102 100 195 k=9 179 174 173 178 178 171 177 171 174 195 k=10 178 172 172 177 178 177 175 173 175 195 k=11 161 152 168 167 167 166 172 157 177 195 k=12 96 106 105 97 101 100 96 104 96 195 k=13 99 101 102 101 99 101 96 102 99 195 k=14 147 155 142 154 146 143 154 153 155 195 k=15 163 152 156 152 153 152 154 156 156 195 k=16 163 185 183 178 169 173 172 182 171 195 k=17 173 179 178 169 173 178 174 175 175 195 k=18 176 167 181 180 172 173 173 168 184 195 k=19 162 150 152 152 160 151 155 148 167 195 k=20 164 161 157 159 162 160 162 158 157 195 k=21 177 161 154 168 172 170 167 149 169 195 Advances in Production Engineering & Management 12(1) 2017 69 Lv, Zhang, Qin Table 6 Production plans in industrial instances Block di d2 d3 d4 ds d6 dy de d9 Pi 30 30 30 30 30 30 30 30 30 P2 30 30 30 45 45 23 23 22 22 P3 10 10 10 60 60 30 30 30 30 P4 40 40 40 15 15 30 30 30 30 P5 40 40 40 60 60 8 8 7 7 P6 50 50 50 30 30 15 15 15 15 P7 20 20 20 75 75 15 15 15 15 P8 20 20 20 30 30 38 38 37 37 P9 70 70 70 15 15 8 8 7 7 Pi0 10 10 10 105 105 8 8 7 7 P11 10 10 10 15 15 53 53 52 52 6 P12 24 23 23 45 45 28 28 27 27 P13 37 37 36 35 35 23 23 22 22 P14 37 37 36 45 45 18 18 17 17 Pi5 24 23 23 55 55 23 23 22 22 P16 30 30 30 35 35 28 28 27 27 P17 30 30 30 55 55 18 18 17 17 P18 60 60 60 30 30 8 8 7 7 P19 10 10 10 90 90 15 15 15 15 P20 20 20 20 15 15 45 45 45 45 P21 60 60 60 15 15 15 15 15 15 P22 20 20 20 90 90 8 8 7 7 P23 10 10 10 30 30 45 45 45 45 P24 60 60 60 60 60 60 60 60 60 P25 60 60 60 90 90 45 45 45 45 P26 20 20 20 120 120 60 60 60 60 P27 80 80 80 30 30 60 60 60 60 P28 80 80 80 120 120 15 15 15 15 P29 100 100 100 60 60 30 30 30 30 P30 40 40 40 150 150 30 30 30 30 P31 40 40 40 60 60 75 75 75 75 P32 140 140 140 30 30 15 15 15 15 P33 20 20 20 210 210 15 15 15 15 P34 20 20 20 30 30 105 105 105 105 7 P35 47 47 40 90 90 55 55 55 55 P36 74 73 73 70 70 45 45 45 45 P37 74 73 73 90 90 35 35 35 35 P38 47 47 40 110 110 45 45 45 45 P39 60 60 60 70 70 55 55 55 55 P40 60 60 60 110 110 35 35 35 35 P41 120 120 120 60 60 15 15 15 15 P42 20 20 20 180 180 30 30 30 30 P43 40 40 40 30 30 90 90 90 90 P44 120 120 120 30 30 30 30 30 30 P45 40 40 40 180 180 15 15 15 15 P46 20 20 20 60 60 90 90 90 90 An Intel(R) Core(TM) i7-2720QM CPU @ 2.20 GHz and 8 GB RAM based notebook computer is used to conduct the computational experiments. Table 7 presents genetic algorithm parameters used in the GRN-based method and the ICGA method. Table 8 lists the average objective function values (Obj) obtained and the average CPU times (Tcpu) spent by the GRN-based method, the ICGA, the HH method and the Cplex solver in each block. In this table, the number of feasible solutions for each instance is evaluated based on Eq. 11 [33]. Nf = (YM dm)!/UM (dm!) (11) dm is the demand for model m in the production plan. The average number of feasible solutions in each block (Nfb) is calculated and also presented in this table as the indicator of problem scales. 70 Advances in Production Engineering & Management 12(1) 2017 A genetic regulatory network-based sequencing method for mixed-model assembly lines Table 7 Genetic algorithm parameters in different methods Method_Population size_Maximum generation_PMutation_PCrossover RCGA 50 30 0.1 0.8 ICGA 200 50 0.1 0.8 Table 8 Experimental results of different blocks Block GRN-based ICGA HH method CPLEX solver Nfb Obj Tcpu, s Obj Tcpu,s Obj Tcpu,s Obj Tcpu, s 1 3.4x103 247.7 0.4 245.2 0.6 248.6 0.5 245.2 4.2 2 8.2x105 136.4 0.5 136.7 0.7 137.1 0.6 135.0 39.3 3 3.8x107 64.5 0.6 68.3 0.9 64.6 0.8 64.3 281.3 4 1.2x107 100.4 0.6 96.5 1.2 100.7 1.1 96.2 190.8 5 5.8x106 114.6 0.6 115.2 0.9 114.7 1.5 113.2 112.2 6 5.8x10247 403.3 107.5 497.2 297.2 419.4 138.1 238410.5 7200 7 7.8x10505 856.0 201.1 924.6 578.4 875.6 216.5 245774.1 7200 As shown in Table 8, based on 'N/b column, two scenarios are considered. Block 1, 2, 3, 4 and 5 are composed of small-scale reference instances and Block 6 and 7 are composed of large-scale industrial instances. In the small-scale reference instances, the Cplex solver obtains the best results, and the other methods obtain results close to the best ones. Specifically, the results in Block 1 reveal that the GRN-based method and the HH method cannot generate the optimal solutions in some instances owing to the predetermined sequencing rules, while the ICGA can obtain the optimal ones through global searching procedure when there are a few feasible solutions. However, the results in Block 2, 3, 4 and 5 reveal that ICGA fails to obtain the optimal solutions for some instances when the number of feasible ones is increased, while the GRN-based method and the HH method generate better solutions than the ICGA. This is because the sequencing rules integrated in the GRN enable the RCGA to search among good solutions rather than all the feasible ones in the regulatory parameter optimization procedure. Similarly, the HH method uses the scatter search to select from different combinations of sequencing rules and thus searches among good solutions too. However, the GRN-based method achieves better results because its weighted integration of commonly-used sequencing rules enables better searching capacity than the random combination of 20 priority rules in the HH method. For the large-scale instances, the Cplex solver fails to obtain good results in the limited CPU time, while the other methods achieve better ones in a reasonable time. The results in Block 6 and 7 reveal that the ICGA can hardly find even near-optimal solutions in an enlarged solution space, while the GRN-based method and the HH method are better than the ICGA. In addition, the results also demonstrate the GRN-based method saves the CPU time. In comparison with the ICGA method, the GRN-based method optimizes four regulatory parameters rather than the whole model sequence to decrease computational effort. This regulatory parameter optimization also demonstrates better efficiency than the scatter search on 20 priority rules in the HH method. Fig. 5 illustrates how the computational time of different methods changes with the increase of problem sizes. The CPLEX solver finds out the optimal solution by using a traversal procedure, for which the computational time increases significantly with a larger problem size. The GRN-based method, the HH method and the ICGA are based on random searching procedures that demonstrate lower increasing rates than the CPLEX solver. Moreover, the ICGA searches among all feasible solutions, whereas the GRN-based method and the HH method search among good solutions owing to the predetermined sequencing rules. For this reason, these two methods save the computational time than the ICGA. Consequently, it can be noted that the GRN-based method provides an effective means to solve the MMS problem, especially for large-scale instances. In addition, this method is also potentially applied for MMS problems in the dynamic environment by using the predictive-reactive strategy. In this strategy, the GRN-based method first provides a production plan with the minimum work overload before line production, and gives the reactive schedule within a rolling window (containing 10~20 products) once the predetermined plan is interrupted by dynamic Advances in Production Engineering & Management 12(1) 2017 71 Lv, Zhang, Qin events such as machine failures or processing time variations. Because the small-sized MMS problems are regularly solved within 1 s by using the GRN-based method, this reactive schedule realizes real-time responses for the dynamic events. In this way, by using a predictive schedule to ensure the overall performance and employing reactive schedules to make quick responses, the GRN-based method will realize efficient production in the dynamic environment. 1000 900 800 700 s 600 & 500 400 300 200 100 0 6. Conclusion This paper deals with the MMS problem in assembly lines to minimize work overload at stations. In terms of similarities between MMS and GRN, a novel MMS method based on the GRN is proposed. This method is applied to reference instances as well as industrial instances to validate its effectiveness. A Cplex solver, an ICGA and a HH method are used to benchmark the results. It is demonstrated that the GRN-based method realizes higher solution quality than other methods by integrating the sequencing rules reasonably, especially for large-sized problems. However, due to the regulatory parameter optimization that uses GA, rather than some well-designed mechanisms, the efficiency of this method requires further improvements. Thereupon, we will investigate a new parameter optimization mechanism in our future work. In addition, some other optimization problems also have the validated rules to determine its binary decision variables. The proposed method is thus potentially used for the problems in other areas, including the production scheduling problem in other manufacturing systems, the transportation scheduling problem in logistics industry and the medical device scheduling problem in healthcare industry. In our future work, we will develop new scheduling methods by extending the GRN-based approach to these areas. Acknowledgement This work was supported by the [National Natural Science Foundation of China] under Grant [number 51435009 & number U1537110]. References [1] Moradi, H., Zandieh, M., Mahdavi, I. (2011). Non-dominated ranked genetic algorithm for a multi-objective mixed-model assembly line sequencing problem, International Journal of Production Research, Vol. 49, No. 12, 3479-3499, doi: 10.1080/00207540903433882. [2] Mukund Nilakantan, J., Ponnambalam, S.G. (2015). Robotic U-shaped assembly line balancing using particle swarm optimization, Engineering Optimization, Vol. 48, No. 2, 231-252, doi: 10.1080/0305215X.2014.998664. [3] Lian, K., Zhang, C., Gao, L., Shao, X. (2012). A modified colonial competitive algorithm for the mixed-model U-line balancing and sequencing problem, International Journal of Production Research, Vol. 50, No. 18, 5117-5131, doi: 10.1080/00207543.2011.653453. -GRN-based method -ICGA -HH method -CPLEX lg(Nfb) Fig. 5 Computational time with different problem sizes 72 Advances in Production Engineering & Management 12(1) 2017 A genetic regulatory network-based sequencing method for mixed-model assembly lines [4] Akpinar, S., Bayhan G.M. (2014). Performance evaluation of ant colony optimization-based solution strategies on the mixed-model assembly line balancing problem, Engineering Optimization, Vol. 46, No. 6, 842-862, doi: 10.1080/0305215X.2013.806915. [5] Boysen, N., Fliedner, M., Scholl, A. (2009). Production planning of mixed-model assembly lines: Overview and extensions, Production Planning & Control, Vol. 20, No. 5, 455-471, doi: 10.1080/09537280903011626. [6] Chan, F.T.S., Wong, T.C., Chan, L.Y. (2009). An evolutionary algorithm for assembly job shop with part sharing, Computers & Industrial Engineering, Vol. 57, No. 3, 641-651, doi: 10.1016/j.cie.2008.11.017. [7] Boysen, N., Kiel, M., Scholl, A. (2011). Sequencing mixed-model assembly lines to minimise the number of work overload situations, International Journal of Production Research, Vol. 49, No. 16, 4735-4760, doi: 10.1080/ 00207543.2010.507607. [8] Bautista, J., Alfaro, R., Batalla, C. (2015). Modeling and solving the mixed-model sequencing problem to improve productivity, International Journal of Production Economics, Vol. 161, 83-95, doi: 10.1016/j.ijpe.2014.11.018. [9] Boysen, N., Fliedner, M., Scholl, A. (2009). Sequencing mixed-model assembly lines: Survey, classification and model critique, European Journal of Operational Research, Vol. 192, No. 2, 349-373, doi: 10.1016/j.ejor.2007. 09.013. [10] Wong, T.C., Chan, F.T.S., Chan, L.Y. (2009). A resource-constrained assembly job shop scheduling problem with Lot Streaming technique, Computers & Industrial Engineering, Vol. 57, No. 3, 983-995, doi: 10.1016/j.cie.2009.04. 002. [11] McMullen, P.R. (2001). An ant colony optimization approach to addressing a JIT sequencing problem with multiple objectives, Artificial Intelligence in Engineering, Vol. 15, No. 3, 309-317, doi: 10.1016/S0954-1810(01) 00004-8. [12] Ding, F.-Y., He, J. (2008). A heuristic procedure for the automobile assembly-line sequencing problem considering multiple product options, International Journal of Production Research, Vol. 46, No. 20, 5827-5847, doi: 10.1080/ 00207540701381291. [13] Franz, C., Hällgren, E.C., Koberstein A. (2014). Resequencing orders on mixed-model assembly lines: Heuristic approaches to minimise the number of overload situations, International Journal of Production Research, Vol. 52, No. 19, 5823-5840, doi: 10.1080/00207543.2014.918293. [14] Rahimi-Vahed, A.R., Rabbani, M., Tavakkoli-Moghaddam, R., Torabi, S.A., Jolai, F. (2007). A multi-objective scatter search for a mixed-model assembly line sequencing problem, Advanced Engineering Informatics, Vol. 21, No. 1, 85-99, doi: 10.1016/j.aei.2006.09.007. [15] Laili, Y., Tao, F., Zhang, L., Sarker, B.R. (2012). A study of optimal allocation of computing resources in cloud manufacturing systems, The International Journal of Advanced Manufacturing Technology, Vol. 63, No. 5, 671-690, doi: 10.1007/s0Q170-012-3939-0. [16] Manavizadeh, N., Tavakoli, L., Rabbani, M., Jolai, F. (2013). A multi-objective mixed-model assembly line sequencing problem in order to minimize total costs in a Make-To-Order environment, considering order priority, Journal of Manufacturing Systems, Vol. 32 No. 1, 124-137, doi: 10.1016/j.jmsy.2012.09.001. [17] Cano-Belmán, J., Ríos-Mercado, R.Z., Bautista, J. (2010). A scatter search based hyper-heuristic for sequencing a mixed-model assembly line, Journal of Heuristics, Vol. 16, No. 6, 749-770, doi: 10.1007/s10732-009-9118-2. [18] Gujjula, R., Werk, S., Günther, H.-O. (2011). A heuristic based on Vogel's approximation method for sequencing mixed-model assembly lines, International Journal of Production Research, Vol. 49, No. 21, 6451-6468, doi: 10.1080/00207543.2010.527384. [19] Cortez, P.M.C., Costa, A.M. (2015). Sequencing mixed-model assembly lines operating with a heterogeneous workforce, International Journal of Production Research, Vol. 53, No. 11, 3419-3432, doi: 10.1080/00207543. 2014.987881. [20] Akgündüz, O.S., Tunali, S. (2010). An adaptive genetic algorithm approach for the mixed-model assembly line sequencing problem, International Journal of Production Research, Vol. 48, No. 17, 5157-5179, doi: 10.1080/ 00207540903117857. [21] Shao, X., Wang, B., Rao, Y., Gao, L., Xu, C. (2010). Metaheuristic approaches to sequencing mixed-model fabrication/assembly systems with two objectives, The International Journal of Advanced Manufacturing Technology, Vol. 48, No. 9, 1159-1171, doi: 10.1007/s00170-009-2337-8. [22] Wang, Y., Liu, J.H. (2010). Chaotic particle swarm optimization for assembly sequence planning, Robotics and Computer-Integrated Manufacturing, Vol. 26, No. 2, 212-222, doi: 10.1016/j.rcim.2009.05.003. [23] Zhu, Q., Zhang, J. (2011). Ant colony optimisation with elitist ant for sequencing problem in a mixed model assembly line, International Journal of Production Research, Vol. 49, No. 15, 4605-4626, doi: 10.1080/00207543. 2010.493534. [24] Bautista, J., Cano, A., Alfaro, R. (2012). Modeling and solving a variant of the mixed-model sequencing problem with work overload minimisation and regularity constraints. An application in Nissan's Barcelona Plant, Expert Systems with Applications, Vol. 39, No. 12, 11001-11010, doi: 10.1016/j.eswa.2012.03.024. [25] Hamzadayi, A., Yildiz, G. (2013). A simulated annealing algorithm based approach for balancing and sequencing of mixed-model U-lines, Computers & Industrial Engineering, Vol. 66, No. 4, 1070-1084, doi: 10.1016/j.cie.2013. 08.008. [26] Pal, R., Bhattacharya, S., Caglar, M.U. (2012). Robust approaches for genetic regulatory network modeling and intervention: a review of recent advances, IEEE Signal Processing Magazine, Vol. 29, No. 1, 66-76, doi: 10.1109/ MSP.2011.942818. [27] Balasubramaniam, P., Banu, L.J. (2013). Robust state estimation for discrete-time genetic regulatory network with random delays, Neurocomputing, Vol. 122, 349-369, doi: 10.1016/j.neucom.2013.06.015. Advances in Production Engineering & Management 12(1) 2017 73 Lv, Zhang, Qin [28] Sun, Y., Feng, G., Cao, J. (2009). Stochastic stability of Markovian switching genetic regulatory networks, Physics Letters A, Vol. 373, No. 18-19, 1646-1652, doi: 10.1016/j.physleta.2009.03.017. [29] Diläo, R., Muraro, D. (2010). Calibration and validation of a genetic regulatory network model describing the production of the protein Hunchback in Drosophila early development, Comptes Rendus - Biologies, Vol. 333, No. 11-12, 779-788, doi: 10.1016/j.crvi.2010.09.003. [30] Dörmer, J., Günther, H.-O., Gujjula, R. (2015). Master production scheduling and sequencing at mixed-model assembly lines in the automotive industry, Flexible Services and Manufacturing Journal, Vol. 27, No. 1, 1-29, doi: 10.1007/s10696-013-9173-8. [31] Bautista, J., Cano, J. (2008). Minimizing work overload in mixed-model assembly lines, International Journal of Production Economics, Vol. 112, No. 1, 177-191, doi: 10.1016/j.ijpe.2006.08.019. [32] Bautista, J., Cano, A. (2011). Solving mixed model sequencing problem in assembly lines with serial workstations with work overload minimisation and interruption rules, European Journal of Operational Research, Vol. 210, No. 3, 495-513, doi: 10.1016/j.ejor.2010.10.022. [33] Tavakkoli-Moghaddam, R., Rahimi-Vahed, A.R. (2006). Multi-criteria sequencing problem for a mixed-model assembly line in a JIT production system, Applied Mathematics and Computation, Vol. 181, No. 2, 1471-1481, doi: 10.1016/j.amc.2006.02.033. [34] Wong, C.S., Chan, F.T.S., Chung, S.H. (2012). A genetic algorithm approach for production scheduling with mould maintenance consideration, International Journal of Production Research, Vol. 50, No. 20, 5683-5697, doi: 10.1080/00207543.2011.613868. [35] Shao, X., Li, X., Gao, L., Zhang, C. (2009). Integration of process planning and scheduling - A modified genetic algorithm-based approach, Computers & Operations Research, Vol. 36, No. 6, 2082-2096, doi: 10.1016/j.cor.2008. 07.006. 74 Advances in Production Engineering & Management 12(1) 2017 APEM jowatal Advances in Production Engineering & Management Volume 12 | Number 1 | March 2017 | pp 75-87 https://doi.Org/10.14743/apem2017.1.241 ISSN 1854-6250 Journal home: apem-journal.org Original scientific paper Influence of the production fluctuation on the process energy intensity in iron and steel industry Chen, D.a, Lu, B.b*, Chen, G.c, Yu, W.b aSchool of Civil Engineering, Anhui University of Technology, Ma'anshan, China bSchool of Energy and Environment, Southeast University, Nanjing, China cSchool of Energy and Environment, Anhui University of Technology, Ma'anshan, China A B S T R A C T A R T I C L E I N F O This paper mainly studies how the production fluctuation affects the process energy intensity in iron and steel industry. First of all, the production state is divided into five conditions according to the production volatility. Meanwhile, the process energy intensity model is constructed. And model analysis showed that operating rate and qualification rate are two key parameters that represent the production volatility. A case study showed that the process energy intensity is inversely proportional to the normal production operating rate and qualification rate, but proportional to the operating rate in the other production states. Moreover, the production halt operating rate and normal production qualification rate significantly influence the process energy intensity in terms of production volatility. And then, some management suggestions were introduced on how to reduce the fluctuation of the process production. The application of the model is quantitative analysis methods, which can describe influence of production fluctuation on the process energy intensity. Based on this, corresponding measures are adopted for reducing energy consumption, including adjustment of production planning and strategy etc. © 2017 PEI, University of Maribor. All rights reserved. Keywords: Iron and steel industry Production fluctuation Production state Operating rate Qualification rate Process energy intensity *Corresponding author: 10793192@qq.com (Lu, B.) Article history: Received 24 June 2016 Revised 20 November 2016 Accepted 31 December 2016 1. Introduction The iron and steel industry, which is the basic raw materials industry, influences the operation of national economy directly [1]. The crude steel production of the world increased from 750 Mmt (Million metric ton) in 1996 to 1545 Mmt in 2012. This is mainly because the crude steel production of China is continuously at the top in the world for a dozen years since 1996, which correspondingly enlarges the energy demand. Take Chinese steelmakers, the energy consumption increased from 168 Mtce (Million ton coal equivalent) in 2000 to 597 Mtce in 2012. The energy consumption of six energy-intensive sectors- excavating industry, chemicals, non-metallic industry, iron and steel, non-ferrous metals, petroleum - accounts for 71.4 % of total industry energy consumption as shown in Fig. 1. It depicts the ratio of energy consumption in each sector to total energy consumption in 2012, with the iron and steel industry having the largest share at 23.6 %. Heavy energy-consuming of the iron and steel industry will inevitably lead to increasing of energy costs and aggravation of the ecological environment. 1) Energy expense challenges The energy cost of iron and steel making accounts for up to 20 % of the total production costs [2]. Coal and electricity are the primary energy source of the iron and steel industry. Meanwhile, 75 Chen, Lu, Chen, Yu Excavating Industry _ 8.4% Others 28.6% Processing of Petroleum,Coking,Process ing of Nuclear Fuel 7.2°% Smelting and Pressing of Non-ferrous Metals 5.9% Manufacture of Raw Chemical Materials and Chemical Products 14.7% Manufacture of Non-metallic Mineral Products 11.6% Smelting and Pressing of Ferrous Metals 23.6% Fig. 1 Energy consumption ratio of each sector (Date source: CHINA STATISTICAL YEARBOOK 2012) due to governmental macro-control, energy prices of coal and electricity have been increasing dramatically since 1980 in China; so high energy consumption and high energy price raises the energy expense of steel making [3]. So energy cost reduction is a very important factor in reducing the total cost. 2) Ecological challenges In terms of ecological challenges, the first issue is carbon dioxide (CO2) emissions. The energy efficiency has a direct impact on overall energy consumption and related CO2 emissions [4].In 2009, the CO2 emission from the Chinese iron and steel sector amounted to 1.17 billion metric tons, which is 16.29 % of the total Chinese CO2 emissions and is nearly equal to 50 % of the world's steel industry's CO2 emissions [5]. And the CO2 emissions from energy (fuel) consumption accounts for 95 % of the total CO2 emissions by the steel industry, which illustrates the important influence of energy (fuel) consumption on CO2 emissions [6]. Meanwhile, GHG emissions problems of steel industry exist widely in other countries as well [7-9]. Secondly, the pollution problem of the steel industry, such as SO2 and NOx, cannot be ignored either [10-12]. Facing the challenges from globalization, enhancing enterprise competitiveness and efficiency by applying new technology and new management means, improving energy efficiency, and reducing production costs is desperately needed. 2. Literature review Currently, energy conservation research works performed in the steel industry have primarily concentrated on three aspects: equipment process improvement, process optimization, and energy conservation through management. 1) Equipment process improvement Equipment process improvement refers mostly to replacing outdated, low-efficiency equipment by advanced, high-efficiency equipment. Early energy conservation efforts were primarily concentrated on the optimization and alteration of individual equipment. Equipment scale enlargement is one of the important measures in energy conservation. For example, blast furnaces are process equipment that have the most concentrated material and energy flow in the iron and steel industry production process [13]. And through a comparison of a 5576 m3 blast furnace from Shougang Jingtang Steel and a 4080 m3 blast furnace from QianAn steel, Zhu et al. [14] discovered that constructing two 5576 m3 blast furnaces can obtain similar yields as constructing three 4080 m3 blast furnaces; however, the former has clear advantages in terms of investment reduction and energy conservation. Meanwhile, large-scale sintering machines and scale enlargement of a coke oven have low energy consumption and a high technical economic index. 76 Advances in Production Engineering & Management 12(1) 2017 Influence of the production fluctuation on the process energy intensity in iron and steel industry Meanwhile, advanced production technology and equipment can improve energy efficiency. Pulverized coal injection technology (PCI) and continuous casting technology (CCT) are effective energy saving technology. The wide use of advanced combustion equipment [15] and power equipment [16] can promote energy efficient utilization. And the use of the spark plasma sintering (SPS) technology allowed for an energy saving in the order of 90-95% [17]. COREX process displays many energy and ecological advantages [18]. The surface of blast furnace tuyere was cooled by using cooling air instead of cooling water, which could reduce the energy taken away by the cooling water [19]. In addition, Tiago et al. [20] analyzed the feasibility of the biomass energy utilization to the EAF steelmaking process. Moreover, the recovery and utilization of residual energy and heat (RUREH) plays an important role for energy saving and CO2 emission reduction [21-22]. 2) Process optimization As technology continues to progress, room for energy savings through process equipment improvement becomes smaller, and thus, process optimization comes into play. The issues of the steel industry studied in process optimization are the internal process function match and coordinated operation; the goal is to achieve continuous, compact production to reduce energy consumption during the production process. Examples of process optimization include technologies such as increasing the continuous casting ratio, long process flow (blast furnace - converter) to short process flow (electric arc furnace steel - continuous casting short process), continuous thin slab casting and rolling processes (CSPs), and rolled steel billet hot charging technology. Since the 1990's, China's steel industry per ton overall energy consumption declined significantly; among which, 48 % of the energy savings was due to the process restructuring and optimization [23]. Lu et al. [24] studied how the way ferrite flows in the iron and steel production process influences energy consumption and proposed important concepts, such as the base operating energy consumption graph. Thereafter, Chen et al. [25] and Yu et al. [26] used relevant concepts to calculate the influence of material flow structure change on energy consumption intensity in an actual production process and searched for an optimized production organization mode that minimized energy consumptions. Moreover, with the development of Circular Economy, industry ecosystem is gradually concerned. Dong et al. [5] agreed with a great potential for implementing circular economy in steel industry, and mode of future steel enterprises in circular economy society was discussed. Song et al. [27] and Chao et al. [28] thought that industry ecosystem is beneficial to itself, economy and environment. 3) Energy conservation through management Energy waste issues in the operation process of the steel industry brought by "evaporating, emitting, dripping, or leaking" and disorderly management have become increasingly prominent. One of the most promising means of reducing energy consumption and related energy costs is implementing an energy management [29]. Wang et al. [30] noted that the potential of energy management system and industrial energy saving policies by transiting steel industry energy flow from "disorderly" to "orderly" is extremely large. Jean-Christian et al. [31] noted when sound energy management practices are included, the participants assessed the cost-effective energy conservation potential to be 9.7 %, which was 2.4 % higher than the potential for solely adopting cost-effective technologies. Tang et al. [32] thought that energy management of steel industry is basically in the safety and insurance mode of production-oriented, leading to inefficient energy management, high energy consumption and great loss of benefits. Therefore, it is necessary to establish a systematic energy efficiency management system oriented by quality and value of energy. Liu et al. [33] pointed out various environmental managements have been taken by the companies, including certain proactive efforts such as conducting cleaner production audit, pursuing ISO 14001 certification and the implementation of ESAs. Because the steel industry production is influenced by factors, such as market demand, up-downstream process linkage, and equipment operation conditions, the production fluctuation changes greatly [23]; this change greatly influences the process energy intensity. When imple- Advances in Production Engineering & Management 12(1) 2017 77 Chen, Lu, Chen, Yu menting lean energy management, the influence of production volatility on the process energy intensity should be fully considered. However, quantitative research on this aspect has rarely been reported. For this reason, this paper takes "process" as the starting point and explores and extracts the primary factors of influence of production fluctuation on the process energy intensity. Chen et al. [34] put forward there were six states in a production process, and process energy intensity formula was constructed. Combined with (E-P) method of cleaner production [35], production states were re-defined and re-divided taken into account this principle of division in this paper, process energy intensity formula was re-established. And the model is further verified using actual examples. And then, some suggestion and management measures were introduced. 3. Study object With the view of process structure, iron and steel enterprise possesses the characteristics of the up-downstream processes are connected in series, and each unit is connected in parallel in the same work procedure. Therefore, When production fluctuations occurs in a unit, it will inevitably bring the production fluctuations of up-downstream processes or other units of the same procedure, thus affecting the energy intensity. In this paper, a unit in the production process is selected as study object (shown in Fig. 2). And then, the influence of production fluctuation on the process energy intensity is discussed. Meanwhile, measures of reducing production volatility, which can reduce impact on process energy intensity, are proposed under different production disposition. ✓ Study object r----- 1 ¡"1 " i L _ J ¡ / r - -i i r , ; ' j . >, . L—> in / >| i 1----v i ! r-- ! 1 m-i 1 i i 1__i 1 Ö i ■ r - m i _ il - ' i k-1 k k+1 Fig. 2 The relationship between study object and if s up-downstream Fig. 2. is the production process flow diagram. In which, k - the k-th procedure; k-1, k+1 - the up-downstream processes, respectively; j - the j unit of the k-th procedure. 4. Methodology 4.1 Production state description Based on (E-P) method of cleaner production, the steel industry total process energy consumption, process production, and the process energy intensity have the following relationship: E = E0+K^P (1) e = E0/P + K (2) In which, E - total process energy consumption within the statistical cycle, tce (ton coal equivalent); E0 - energy consumption not directly related to production within the statistical cycle, such as energy consumed by the company general service, tce; P - process production within the statistical cycle, t; K- the normal production state energy intensity, tce/t; e - the process energy intensity within the statistical cycle, tce/t. Fig. 3. is the process energy production graph (E-P graph), where the total process energy consumption increases as production increases, whereas the process energy intensity decreases in an inversely proportional manner. When the production is zero (stop production), equipment used for the company general service still consumes energy; therefore, the total process energy 78 Advances in Production Engineering & Management 12(1) 2017 Influence of the production fluctuation on the process energy intensity in iron and steel industry K 0 Production Fig. 3 Relationship between energy consumption and production of process consumption is not zero, and thus, the process energy intensity is infinite in this situation. When the process production is relatively low (abnormal production), despite the total process energy consumption being low, due to the low production amount, the process energy intensity is extremely high. With the gradual increase in production, the total process energy consumption increases correspondingly; however, the process energy intensity decreases significantly; also, when production increases to the designed capacity (normal production), the process energy intensity approaches a stable state. Therefore, this paper divides the process production state into three conditions of stop production, abnormal production, and normal production. Additionally, transitional states must exist between these three states, as shown in Fig. 4. The processes of transitional states (4), (5), and (6) belong to production decrease; the energy consumption is relatively low and thus, will not be considered. Moreover, the occurrence probability of the direction transition from normal production to abnormal production process (process (1) in Fig. 4.) is extremely low and therefore, will not be considered. Transitional states (2) and (3) are processes from low production volume to high production volume, in which the majority of the equipment in the process is at a start-up stage, and the energy consumption is relatively high. So it is adequate to only consider process transitional states (2) and (3). Here, state (2) is defined as a stop production transition, and state (3) is defined as an abnormal production transition. Combined with Fig. 2, the relationship between the total energy consumption and the process energy intensity for each production state can be obtained, as shown in Fig. 5. While, the total process energy consumption in normal production is highest, but the process energy intensity is lowest; the total process energy consumption in stop production is lowest; however, because the production volume is zero, the process energy intensity is infinite. Fig. 4 Illustration of the production state Relative total process energy consumption Relative process energy intensity Relative Process Energy Intensityof Normal Production E0 Advances in Production Engineering & Management 12(1) 2017 79 Chen, Lu, Chen, Yu Relative Energy Consumption | | Relatvie Energy Intensity I I I I Normal Abnormal Stop Abnormal Stop Production Production Production Production Production Transition Transition Fig. 5 Relationship of all production states between the relative energy consumption and relative energy intensity 4.2 Mathematical model The concept of the statistical average of energy and material flow is introduced. Energy flow is the ratio of the total energy flow amount in the /th state and time duration, Shown in Eq. 3. nt ni Ei=Yjj l'lEi,i(t)dt/Yjti,l (3) 1=1 0 1 = 0 In which, / - subscript, / =1, 2, 3, 4, 5, which represents normal production, abnormal production, stop production, abnormal production transition, and stop production transition, respectively; Ei - statistical average of energy flow in /-th production state, tce/h; ni - number of times the /-th production state occurred; Eitl - instantaneous energy flow in the /-th occurrence of /-th state, tce/h; tifi - time of /-th occurrence of /-th state, h. By the same principle, the material flow statistical average can by expressed by Eq. 4: Znt rti,l v—i^i I Mirldt/) (4) i=o JQ ¿—ii=o In which, Mt - statistical average of material flow in /-th production state, t/h; M^ - instantaneous material flow in the /-th occurrence of /-th state, t/h; 1) Total process energy consumpt/on The total process energy consumption is the sum of the energy consumption in each of the five production states within the statistical cycle: E = E1T1 + E2T2 +E3T3 + E4T4 + (5) = T(E1'n1 + E2y 2 +E3y 3 In which, Ti - duration of /-th state, h; T - statistical cycle, h; ^ - operating rate of /-th state, %, it is defined as T\/T. 2) Process product/on The process production in the statistical cycle is the sum of the qualified product in each of the normal production, abnormal production, stop production, abnormal production transition, and stop production transition (volume of stop production is 0) state: P= M1T1ripl +M2T2rlp2 + M4T4Vp4 + M57^p5 (6) = T(M1^1'npl +M2V2VP2 + M4il4ilp4 + ( ) In which, rçpi - qualification rate of /-th state, %. 80 Advances in Production Engineering & Management 12(1) 2017 Influence of the production fluctuation on the process energy intensity in iron and steel industry 3) The process energy intensity The process energy intensity is: " = E/P m Let at = Ei/E1 , fa =Mi/M1; then, the process energy intensity model is Rs • T (/: inventory level, metric ton; Rs: normal supply of upstream production process, metric ton/hour; T:shutdown time, hour). The demand for raw materials of the research object, that is Rs, is actively reduced in order to keep the production stability if / < Rs • T, and research object is in abnormal production state at this moment. From the foregoing analysis, abnormal production is far superior to the stop production in the energy intensity. Where, it can only be discontinued if production decrease still una- 0............ '0 1 1 1 Upstream prodeution process (a) Upstream prodeution process (b) P"" 1 ' 1 ntory !> 1 1 1 L _ 01 Upstream prodeution process (c) Upstream prodeution process (d) Fig. 8 The structure between production processes 84 Advances in Production Engineering & Management 12(1) 2017 Influence of the production fluctuation on the process energy intensity in iron and steel industry ble to ensure continuous production. There are many production units in the upstream production process in Fig. 8(c) and Fig. 8(d), and there are characteristics of parallel between units. The research object will not stop production if one or a few units are discontinued (Not all). So the upstream production process also needs to formulate strict equipment maintenance regulations and reasonable repair schedules to minimize the probability of the occurrence and time of stop production. In addition, the establishment of the buffer unit (that is inventory) is beneficial to reduce the fluctuation of production. 3) The lack of market demand Iron and steel enterprise need to reduce production because the shortage in market need. And the production capacity of iron and steel enterprise can't be fully released. So some measures should be done in order to make the production can be operated smoothly, such as readjusting product structure and remaking production schedule, according to changes in market demand. 6.2 Analysis of qualification rate The main impacts of qualification rate are as follows in practical production. 1) The quality of raw material and fuel The quality of raw materials is critical to the qualification rate of product. The impurity content of iron ore, which is the main raw material for iron and steel enterprise, will directly affect the quality of molten iron; even affect product quality of the subsequent process. For example, Sulfur and phosphorus are typical harmful elements in iron ores. High content of sulfur will cause the steel hot brittleness; reduce the ductility and toughness of the steel, and cracks are formed in forging and rolling. Meanwhile, sulfur is also detrimental to the welding performance and reduces corrosion resistance. In addition, High content of phosphorus will increase the cold brittle-ness of steel, deteriorate the welding performance and cold-bending property, and reduce the plasticity. The influence of fuel quality can't be ignored. For instance, the poor permeability of blast furnace and the furnace condition stability are affected when ash and sulfur content increase. Moreover, COG (Coke oven gas) is the major source of energy for heating furnace, some performance of steel will be poor, such as hot brittleness, if it contains high sulfur content. 2) Operating parameters of production process Operating rules and regulations must be set up to strictly control the parameters in order to guarantee the quality of product. Otherwise, scrap rate will increase, and even cause stop production. For example, the blast kinetic energy determines the size and shape of the combustion zone in the furnace. If the blast kinetic energy is small, the gas distributes in the edge area. And conversely the center of the gas flow is disturbed. These two kinds of conditions can lead to poor quality of molten iron. Meanwhile, the billet is easy to produce surface crack and columnar crystal if the pouring temperature of the continuous casting machine is too high. Conversely inclusion can't float. Moreover, the billet is over heated and oxide scale of steel will increase if the heating furnace gas flow is too large. Conversely the billet can't be fully heated and can't be rolled. 3) Aging of equipment or backward production techniques Inferior efficiency will happen due to aging of equipment or backward production techniques. Furthermore, qualification rate will reduce. So aging equipment and backward production techniques should be replaced or eliminated promptly. 7. Conclusions and suggestions According to above-mentioned analysis, the following achievement can be obtained: • This paper divides the process production state into five conditions: normal production, abnormal production, stop production, abnormal production transition, and stop production transition through analysing (E-P) method of cleaner production; and then, the process energy intensity model is constructed. Advances in Production Engineering & Management 12(1) 2017 85 Chen, Lu, Chen, Yu • Operating rate and qualification rate (especially the stop production operating rate and normal production qualification rate) are important index of production fluctuation on process energy intensity through case study. Meanwhile, operating rate and qualification rate are analysed in order to reduce the impact of production volatility on process energy intensity. And some suggestions are proposed. • Rules and regulations is a prerequisite to ensure the normal operation of equipment, and consummate rules should be formulated, such as daily inspection tour, spot inspection and fault treatment plans and so on. In addition, on-line fault diagnosis system also helps to reduce the probability of failure of equipment. • Operation and maintenance personnel are the executive of the rules and regulations, so they need to be trained and assessed regularly to improve vocational skills. • Buffer unit, which is conducive to reducing the volatility of production, should be adopted between production processes in iron and steel enterprise. • The high-grade of raw/fuel and advanced production techniques favors to improve qualification rate of the product. • Readjusting product structure and remaking production schedule can avoid production halt or abnormal production according to changes in market demand. Acknowledgements This work is supported by The National Key Technology R&D Program during the 12th Five-Year Plan Period (2012BAF10B05). Reference [1] Du, T., Shi, T., Liu, Y., Ye, J.-B. (2013). Energy consumption and its influencing factors of iron and steel enterprise, Journal of Iron and Steel Research, International, Vol. 20, No. 8, 8-13, doi: 10.1016/S1006-706X(13)60134-X. [2] Rasul, M.G., Tanty, B.S., Mohanty, B. (2007). Modelling and analysis of blast furnace performance for efficient utilization of energy, Applied Thermal Engineering, Vol. 27, No. 1, 78-88, doi: 10.1016/j.applthermaleng. 2006. 04.026. [3] Zheng, L. (2012). A system dynamics based study of polocies on reducing energy use and energy expense for Chinese steel industry, Foreign Investment in China, No. 8, 156-157. [4] Hasanbeigi, A., Price, L., Chunxia, Z., Aden, N., Xiuping, L., Fangqin, S. (2014). Comparison of iron and steel production energy use and energy intensity in China and the U.S, Journal of Cleaner Production, Vol. 65, 108-119, doi: 10.1016/j.jclepro.2013.09.047. [5] Dong, L., Zhang, H., Fujita, T., Ohnishi, S., Li, H., Fujii, M., Dong, H. (2013). Environmental and economic gain of industrial symbiosis for Chinese iron/steel industry: Kawasaki's experience and practice in Liuzhou an Jinan, Journal of Cleaner Production, Vol. 59, 226-238, doi: 10.1016/j.jclepro.2013.06.048. [6] Zhang, C.-X., Shangguan, F.-Q., Hu, C.-Q., Qi, Y.-H., Yin, R.-Y. (2010). Steel process structure and its impact on CO2 emission, Iron and Steel, Vol. 45, No. 5, 1-6, doi: 10.13228/j.boyuan.issn0449-749x.2010.05.021. [7] Fysikopoulos, A., Papacharalampopoulos, A., Pastras, G., Stavropoulos, P., Chryssolouris, G. (2013). Energy efficiency of manufacturing processes: A critical review, Procedia CIRP 7 - Forty Sixth CIRP Conference on Manufacturing Systems 2013, Vol. 7, 628-633, doi: 10.1016/j.procir.2013.06.044. [8] Worrell, E., Price, L., Martin, N. (2001). Energy efficiency and carbon dioxide emissions reduction opportunities in the US iron and steel sector, Energy, Vol. 26, No. 5, 513-536, doi: 10.1016/S0360-5442(01)00017-2. [9] Arens, M., Worrell, E., Schleich, J. (2012). Energy intensity development of the German iron and steel industry between 1991 and 2007, Energy, Vol. 45, No. 1, 786-797, doi: 10.1016/j.energy.2012.07.012. [10] Li, Z.-P., Fan, X.-H., Yang, G.-M., Wei, J.-C., Sun, Y., Wang, M. (2015), Life cycle assessment of iron ore sintering process, Journal of Iron and Steel Research, International, Vol. 22, No. 6, 473-477, doi: 10.1016/S1006-706X(15) 30029-7. [11] Zhou, H., Cheng, M., Zhou, M., Liu, Z., Liu, R., Cen, K. (2016). Influence of sintering parameters of different sintering layers on NOx emission in iron ore sintering process, Applied Thermal Engineering, Vol. 94, 786-798, doi: 10.1016 /j.applthermaleng.2015.09.059. [12] Zhang, S., Worrell, E., Crijns-Graus, W., Wagner, F., Cofala, J. (2014). Co-benefits of energy efficiency improvement and air pollution abatement in the Chinese iron and steel industry, Energy, Vol. 78, 333-345, doi: 10.1016/ j.energy.2014.10.018. [13] Liu, X., Chen, L., Qin, X., Sun, F. (2015). Exergy loss minimization for a blast furnace with comparative analyses for energy flows and exergy flows, Energy, Vol. 93, Part 1, 10-19, doi: 10.1016/j.energy.2015.09.008. 86 Advances in Production Engineering & Management 12(1) 2017 Influence of the production fluctuation on the process energy intensity in iron and steel industry [14] Zhu, R., Zhu, J., Li, J. (2010). Development of and exploration on large-scale blast furnaces, World ¡ron & Steel, Vol. 10, No. 5, 33-39, doi: 10.3969/j.issn.1672-9587.2010.05.007. [15] Zhang, F.-M., Mao, Q.-W., Mei, C.-H., Li, X., Hu, Z.-R. (2012). Dome combustion hot blast stove for huge blast furnace, Journal of ¡ron and Steel Research, International, Vol. 19, No. 9, 1-7, doi: 10.1016/S1006-706X(13)60001-1. [16] Napp, T.A., Gambhir, A., Hills, T.P., Florin, N., Fennell, P.S. (2014). A review of the technologies, economics and policy instruments for decarbonising energy-intensive manufacturing industries, Renewable and Sustainable Energy Reviews, Vol. 30, 616-640, doi: 10.1016/j.rser.2013.10.036. [17] Musa, C., Licheri, R., Locci, A.M., Orru, R., Cao, G., Rodriguez, M.A., Jaworska, L. (2009). Energy efficiency during conventional and novel sintering processes: The case of Ti-AhO3-TiC composites, Journal of Cleaner Production, Vol. 17, No. 9, 877-822, doi: 10.1016/j.jclepro.2009.01.012. [18] Ziebik, A., Lampert, K., Szega, M. (2008). Energy analysis of a blast-furnace system operating with the Corex process and CO2 removal, Energy, Vol. 33, No. 2, 199-205, doi: 10.1016/j.energy.2007.09.003. [19] Shen, Y.S., Zong-Ming, L., Tao, Z., Fu-Sheng, Y., Hong-Ni, X., Rui-Lian, S. (2009). The new technology and the partial thermotechnical computation for air-cooled blast furnace tuyere, Applied Thermal Engineering, Vol. 29, No. 5-6, 1232-1238, doi: 10.1016/j.applthermaleng.2008.06.026. [20] Oliveira, T.L., Assis, P.S., Leal, E.M., Ilídio, J.R. (2015). Study of biomass applied to a cogeneration system: A steelmaking industry case, Applied Thermal Engineering, Vol. 80, 269-278, doi: 10.1016/j.applthermaleng.2015. 01.002. [21] Chen, L., Yang, B., Shen, X., Xie, Z., Sun, F. (2015). Thermodynamic optimization opportunities for the recovery and utilization of residual energy and heat in China's iron and steel industry: A case study, Applied Thermal Engineering, Vol. 86, 151-160, doi: 10.1016/j.applthermaleng.2015.04.026. [22] Walsh, C., Thornley, P. (2012). Barriers to improving energy efficiency within the process industries with a focus on low grade heat utilization, Journal of Cleaner Production, Vol. 23, No. 1, 138-146, doi: 10.1016/j.jclepro.2011. 10.038. [23] Yin, R. (2011). Metallurgical process engineering, Springer, Beijing, Metallurgical Industry Press, Beijing, China, doi: 10.1007/978-3-642-13956-7. [24] Lu, Z., Cai, J., Yu, Q., Xie, A. (2000). The influences of materials flows in steel manufacturing process on its energy intensity, Acta Metallurgica Sinica, Vol. 36, No. 4, 370-378, doi: 10.3321/j.issn:0412-1961.2000.04.008. [25] Chen, G., Cai, J.-J., Yu, Q.-B., Lu, Z.-W. (2002). The analysis of the influences of materials flows in iron and steel corporation on its energy consumption, Journal of Northeastern University (Natural Science), Vol. 23, No. 5, 459462, doi: 10.3321/j.issn:1005-3026.2002.05.014. [26] Yu, Q.-B., Lu, Z.-W., Cai, J.-J. (2007). Calculating method for influence of material flow on energy consumption in steel manufacturing process, Journal of ¡ron and Steel Research, International, Vol. 14, No. 2, 46-51, doi: 10.1016/ S1006-706X(07)60026-0. [27] Chae, S.H., Kim, S.H., Yoon, S.-G., Park, S. (2010). Optimization of a waste heat utilization network in an eco-industrial park, Applied Energy, Vol. 87, No. 6, 1978-1988, doi: 10.1016/j.apenergy.2009.12.003. [28] Gu, C., Leveneur, S., Estel, L., Yassine, A. (2013). Modeling and optimization of material/energy flow exchanges in an eco-industrial park, Energy Procedia, Vol. 36, 243-252, doi: 10.1016/j.egypro.2013.07.028. [29] Schulze, M., Nehler, H., Ottosson, M., Thollander, P. (2016). Energy management in industry - A systematic review of previous findings and an integrative conceptual framework, Journal of Cleaner Production, Vol. 112, Part 5, 3692-3708, doi: 10.1016/j.jclepro.2015.06.060. [30] Wang, Y., Li, H., Song, Q., Qi, Y. (2015). The consequence of energy policies in China: A case study of the iron and steel sector, Resources, Conservation and Recycling, Vol. 117, Part A, 66-73, doi:10.1016/j.resconrec.2015.07.007. [31] Brunke, J.-C., Johansson, M., Thollander, P. (2014). Empirical investigation of barriers and drivers to the adoption of energy conservation measures, energy management practices and energy services in the Swedish iron and steel industry, Journal of Cleaner Production, Vol. 84, 509-525, doi: 10.1016/j.jclepro.2014.04.078. [32] Tang, E., Shao, Y.-J., Fan, X.-G., Ye, L.-D., Wang, J. (2014). Application of energy efficiency optimization technology in steel industry, Journal of ¡ron and Steel Research, International, Vol. 21, Supplement 1, 82-86, doi: 10.1016/ S1006-706X(14)60126-6. [33] Liu, X., Niu, D., Bao, C., Suk, S., Shishime, T. (2012). A survey study of energy saving activities of industrial companies in Taicang, China, Journal of Cleaner Production, Vol. 26, 79-89, doi: 10.1016/j.jclepro.2011.12.030. [34] Chen, G. (2004). Development of process energy intensity formula under different state variables, Journal of Harbin Institute of Technology (NewSeries), Vol. 11, No. 6, 694-696, doi: 10.3969/j.issn.1005-9113.2004.06.025. [35] Yang, L. (2009). Assessment approaches to cleaner production audit with resource and energy as auditing keynote, Environmental Science and Management, Vol. 34, No. 7, 153-156, doi: 10.3969/j.issn. 1673-1212.2009. 07.044. Advances in Production Engineering & Management 12(1) 2017 87 Advances in Production Engineering & Management Volume 12 | Number 1 | March 2017 | pp 88-96 https://doi.Org/10.14743/apem2017.1.242 ISSN 1854-625G Journal home: apem-journal.org Original scientific paper Automatic high resolution measurement set-up for calibrating precise line scales Klobucar, R.a*, Acko, B.a aFaculty of mechanical engineering, University of Maribor, Maribor, Slovenia A B S T R A C T A R T I C L E I N F O This paper presents a high resolution measurement set-up developed for calibrating precise line scales with measurement uncertainty of less than 0.1 |m over a total length of 500 mm. The system integrates a numerically controlled multi-axis stage, a laser interferometer, and a vision system for detecting line position. The measurement and the analysis processes are completely automated in order to minimize manual labour during the calibration process, but also increase the calibration accuracy. Increasing calibration accuracy leads up to better quality of industrial measurements which is required by modern precision industry. When designing this set-up, special attention was paid to the alignment of the measurement object in the measurement direction, considering the focus of the camera. The aim of this alignment was to reduce Abbe errors in 2 axes to negligible level. In addition, all uncertainty contributions have been determined and evaluated by performing extended experiments in specific measurement conditions. These contributions are presented in the uncertainty budget. The metrological capabilities of the presented measurement set-up were verified by some practical test measurements. Selected results of these measurements are presented in the article. This set-up will primarily improve a standard base for calibration of optical measuring devices. The use of the optical standards in the industry is constantly growing. Indirect users of the results of this research will be all manufacturers of precise products such as automotive and other industries. Keywords: Measurement Line scales High resolution measurements Measurement uncertainty *Corresponding author: rok.klobucar@um.si (Klobucar, R.) Article history: Received 15 September 2016 Revised 27 February 2017 Accepted 1 March 2017 © 2017 PEI, University of Maribor. All rights reserved. 1. Introduction Line scales and grid plates are most common measurement standards for assuring traceability of optical measuring equipment, such as microscopes, profile projectors and digital "vision" systems. They are widely used in industry and research laboratories as calibration standards [8]. Line scales are important material standards of length, used for accurate positioning or measurement in one, two or three dimensions [1]. Industry is demanding more and more rapid and accurate dimensional measurements on diverse mechanical parts. In recent years, optical coordinate measuring machines (CMMs) having made substantial progress and are now often used for such applications. Especially CMMs equipped with imaging capabilities are frequently used for fast, non-contact measurements. Strong competition among manufacturers of such instruments and the demand for sub-micron accuracy has led to standardised tests, which are aimed for comparing and validating instrument performance [2]. Calibration, verification and also error correction of 88 Automatic high resolution measurement set-up for calibrating precise line scales optical CMMs are mainly based on measurements using reference line scales or two-dimensional grid plates [3]. There are very many commercially available line scales made of different materials (steel, brass, invar, glass, quartz, zerodur). They can vary in length from below 1 mm to more than 1 m, and have resolutions (pitch) from below 10 nm to 1 cm. They can be calibrated by using different measurement set-ups, depending on their length and precision. Set-ups for calibrating high-precision line scales normally involve a microscope with an optical sensor for capturing and analysing the image of a line marker and a laser interferometer as a traceable measurement standard [4-6]. This paper introduces a measuring system developed for calibrating line scales with measurement uncertainty of less than 10 nm over a total length of 500 mm. The system integrates a numerically controlled multi-axis stage, a laser interferometer, and a vision system for detecting line position. With this measurement set-up, Abbe errors can be reduced to negligible levels. Abbe error occurs when the measurement point of interest is spatially separated laterally from the actual measuring scale location (reference line or axis of measurement), and when angular error motions exist in the positioning system. Abbe error causes measured displacement to appear longer or shorter than the true displacement, depending on the direction of angular motion. Spatial separation between the measured point and reference line is known as an Abbe offset [7]. Since it is possible to put optical components very close to each other, the air dead path error of the laser interferometer is also negligible. Software for detecting the middle of the lines is based on earlier design from 2009 [9]. In order to improve performance and to achieve better uncertainty, the software was improved and automatized, mechanics and optics have been redesigned and several uncertainty components better characterized. This facility will replace the old measurement facility that is manually operated. A model for evaluating uncertainty and the uncertainty budget for the demonstrated measuring system is presented in the paper. 2. Measurement system 2.1 Numerically controlled multi-axis stage Numerically controlled multi-axis stage was designed and manufactured by Newport-micro controle [10] for the Laboratory for Production Measurement (LTM), University of Maribor, Slovenia. It is shown in Fig. 1, while its schematic diagram is shown in Fig. 2. The measuring system designed by the LTM is intended to perform 2D measurements on various objects, which will be fixed on a measuring table. This research is focused in 1D measurements. Newport's solution is based on a HybrYX-G5 stage featuring a ceramic carriage which freely slides in X and Y axes on a precision lapped granite reference plate, using proprietary pressure-vacuum air bearing design. The carriage is guided along Y axis by a rigid ceramic beam. The beam is supported and guided at each end with the ball bearing carriages of X axis. Both X and Y axes are motorized with linear actuators and include linear glass encoders. Both linear scale glass encoders are LIDA403 made by Heidenhain, length 440 mm for X axis and length 1250 mm for Y axis. This positioning system is equipped with the optional Z-Tip-Tilt and Theta stage. It is motorized by precision actuators equipped with miniature DC servo motors. The position system is built on a heavy granite table, which features a precision reference plane for the air bearing carriage. In addition, the granite structure includes a granite gantry allowing the fixation of a vertical motorized translation stage, which accommodates the measuring sensor, depending on an application. Two rails mounted on the gantry equipped with sliding carriages provide manual rough positioning capability. The granite structure is set on a heavy welded frame equipped with ND40 passive isolators. The position system has one XPS motion controller including dedicated drivers for the five DC motors and the three linear motors. The X axis is dual axis composed of X1 and X2 actuators mounted on the flanks at both ends of the granite table. X1 and X2 carriages move the ceramic beam of the Y axes. The X axis travel Advances in Production Engineering & Management 12(1) 2017 89 Klobučar, Acko range is +/- 175 mm. It has a linear encoder with 5 nm resolution. The maximal speed is 300 mm/s., while the repeatability is 500 nm. The Y axis is a horizontal translation stage using air bearing technology and linear motor. The guiding beam of Y axes sub-assembly is attached to the carriages of the X1 and X2 short travel translation actuators. The guideways involve two parts: a large ceramic carriage and a Y-axis guide ceramic L-shape beam. The ceramic carriage freely slides in X and Y axes using pressure-vacuum air bearing design. The reference planes for the air bearing carriage are the precision lapped surface of the granite table and the vertical slide of the Y rigid ceramic beam. The Y axis travel range is +/- 500 mm. It contains a linear encoder with 5 nm resolution. Maximum speed is 600 mm/s., while the repeatability is 100 nm. The ZTT Theta stage is mounted on the air bearing carriage; it is equipped at the top with the measuring platform, which allows installing measured objects. All axes are driven by the high precision motorized linear actuators. Three actuators are set in upright position with a travel range of 10 mm. One actuator is set horizontally to achieve the theta Z movement with a travel range of +/- 1°. All four actuators are equipped with a miniature DC motor made by Fulhaber. Z axis is a vertical motorized translation stage on the middle carrier. It is a Newport catalogue stage reference M-IMS100V. The Z axis travel range is 100 mm, the minimum incremental motion is 0.3 ^.m, the encoder resolution is 0.1 ^m, the maximum speed is 20 mm/s, while the repeatability is ± 0.5 ^.m. Fig. 1 Measurement set-up for calibrating line scales Measuring beam LASER HEAD Fig. 2 Schematic diagram of measurement set-up for calibrating line scales 90 Advances in Production Engineering & Management 12(1) 2017 Automatic high resolution measurement set-up for calibrating precise line scales 2.2 Laser interferometer Laser interferometer position measurement systems provide very precise position or distance information. Our system consists of a laser head HP 5528A, Agilent module 55292A, and a variety of optical components and accessories such as material sensors and air sensor. The basic system measures linear displacement The system uses the wavelength of light from a low-power helium-neon laser as a length standard. We normally set the resolution of the laser interferometer to 10 nm. Special mounting elements were constructed for the optics used for calibration. A schematic diagram of the laser interferometer and the position of the optics are shown in Fig. 2. The position of the optical elements and the moving parts are set in such way, that the measured object axis is set in the line with the centre of the laser linear retroreflector. It allows the Abbe errors to be reduced to negligible levels even for the most demanding dimensional metrology tasks [11]. The linear interferometer is fixed on the fixed part of the stage, while the retroreflector is placed on the moving table in the horizontal direction. Only the measured object is placed on the moving table that can move in the vertical direction. 2.3 Vision system for detecting line position The vision system for detecting line position consists of a zoom microscope and a CMOS digital camera. The camera is connected to the computer via USB 3.0 port. The CMOS camera gets the images of the line scale and sends them to the computer software. The software analyses the images and determinates the middle of the line in the measured window as shown in Fig. 3, which is defended by the operator. The software calculates the distance in pixels from the reference position, which is marked with the blue line, to the middle of the measured line that, is marked with the red line. The image positioning screen with the measurement window, reference and measured line is presented in Fig. 3. The CMOS digital camera takes 15 monochrome images per second in resolution 2592 x 1944 pixels. The software analyses the images in real time. The distance calculated in pixels is transformed into micrometers [12]. The software for calculation the distance between lines is more detailed presented by the authors in paper [9]. Fig. 3 Screen image of the vision system for detecting line position 2.4 Measurement procedure Line scales are material measures made of glass, steel or other material, on which dimensions are marked with line marks. Since the materials have quite different temperature expansion coefficients, which are in many cases notexactly known, they are stabilized before calibration in the climatic room at 20 °C for 24 hours. The measurement system must be adjusted and initialized before the measurement The line scale should be positioned under the camera by moving the measurement table. The measuring system, laser interferometer, vision system for detecting line position and line scale should be adjusted according to the measuring direction, which is defended by the movement of the horizontal direction of the table. The line scale should be fixed on the adjustment table under the camera. The camera is fixed. It is adjusted in such a way, that the focus of the camera intersects the centre of the laser linear retroreflector. The camera is focused on the lines of the line scale by moving the adjustment table (ZTT Theta stage) Advances in Production Engineering & Management 12(1) 2017 91 Klobučar, Acko in the vertical direction. With this procedure, the line scale axis, the images of the lines and the centre of the laser linear retroreflector with the laser beam are positioned in the measuring direction. The image processing software should be initialized [9]. With the improved software, connection to the numerically controlled multi-axis stage and connection to the laser head, automatic measurements of distances are possible. The measurement goal is to measure distances between the reference line and the chosen measuring lines on the line scale automatically. Measured data are saved into the file where they are ready for further processing. 2.5 Principle for minimising Abbe error In respect to the movement parts of the measuring system, the system can be divided into three different parts. The first part is the fixed part ant it includes the laser head, the linear interferometer and the USB 3.0 camera. The second part is the moving part in the horizontal direction and it includes the linear retroreflector, the moving table and the measured object. The third part is the moving part in the vertical direction and it includes the moving table and the measured object The third part is used only for adjusting the line scale in respect to the measuring object direction. The camera is in the fixed distance, so that the focus of the camera is on the top of the moving table. This means that the theoretical measured lines that could be on the top of the table, are in the focus of the camera and simultaneously with the laser optics. Only the moving table with the line scale could be moved in the vertical direction. This procedure allows us moving the lines of the line scale into the focus of the camera and in the same time into the centre of the laser optics. With this procedure, the minimum Abbe error can be achieved. One of the most important sources of error in dimensional measurement is Abbe error. Abbe error consists of an Abbe offset and a small angular variation due to pitch and yaw of the stage that moves the measuring table with the measured object and the linear retroreflector. This component is caused by the measuring table inclination along the measurement path (Y axis). The best available tool for analysing angular error is a laser interferometer. Pitch and yaw measurements were made by making angular measurements at multiple points along the linear travel path in the Y axis as shown in Fig. 4. The angles of the measuring table were measured by using angular interferometer optics. The greatest Abbe errors were evaluated by analysing extreme angular deviations. For pitch (Fig. 4), the greatest angle differences along the measurement path of 500 mm was 7 ^m/m. The maximum offset of the laser reflector of 3 mm in the Z direction was estimated. Therefore the expected error interval for pitch is: 1 = 7 nm/mm • 3 mm = 21 nm (1) Standard uncertainty at supposed rectangular distribution for the Abbe error for pitch is: Fig. 4 Pitch and yaw measurements of the stage in Y direction 92 Advances in Production Engineering & Management 12(1) 2017 Automatic high resolution measurement set-up for calibrating precise line scales For yaw (Fig. 4), the greatest angle differences along the measurement path of 500 mm was 10 |j.m/m. The maximum offset of the laser reflector of 1 mm in the X direction was estimated. Therefore the expected error interval for the yaw is: / = 10 nm/mm • 1 mm = 10 nm (3) Standard uncertainty at supposed rectangular distribution for the Abbe error for the yaw is: 10 nm Uxy(