A Comparison of Three Search Algorithms for Solving the Buffer Allocation Problem in Reliable Production Lines L. Demir 1 , A. Diamantidis 2 , D.T. Eliiyi 3 , M.E.J. O’Kelly 4 , C.T. Papadopoulos 2,# , A.K. Tsadiras 2 , S. Tunalı 5 # Corresponding author: hpap@econ.auth.gr 1 Department of Industrial Engineering, Pamukkale University, Kinikli Campus, Denizli 20070, Turkey (email: ldemir@pau.edu.tr) 2 Department of Economics, Aristotle University of Thessaloniki, Greece (e-mails: adiama@econ.auth.gr, hpap@econ.auth.gr, tsadiras@econ.auth.gr) 3 Department of Industrial Systems Engineering, Izmir University of Economics, Sakarya Cad. No: 156, Balcova-Izmir, Turkey (email: deniz.eliiyi@ieu.edu.tr) 4 Waterford Institute of Technology, Waterford, Ireland (e-mail: mejokelly@gmail.com) 5 Department of Business Administration, Izmir University of Economics, Sakarya Cad. No: 156, Balcova-Izmir, Turkey (email: semra.tunali@ieu.edu.tr) Abstract: This paper investigates the performance of three search algorithms: Myopic Algorithm, Adaptive Tabu Search and Degraded Ceiling to solve the buffer allocation problem in reliable production lines. DECO algorithm is used to calculate throughput. This algorithm is a variant of a decomposition algorithm specifically developed to solve large reliable production lines with parallel machines at each workstation and exponentially distributed service times. The measures of performance used are the CPU time required and closeness to the maximum throughput achieved. The three search algorithms are ranked in respect to these two measures and certain findings regarding their performances over the experimental set are given. Keywords: Production lines, Design, Optimization problems, Buffer storage, Search methods, Algorithms 1. INTRODUCTION & LITERATURE REVIEW ‘Production systems are complex but not evil’ was stated by Li and Meerkov (2009) paraphrasing the well-known Einstein’s ‘nature is complex but not evil’. There are many issues involved in the topology, structure, analysis, design and operation of manufacturing systems. In this paper, we deal with a design problem in serial production lines, known as the buffer allocation problem (BAP). This problem is concerned with the specification of the sizes of the buffers between stations. The provision of buffer space involves considerable cost but leads to increased throughput. There is an extensive bibliography in the area of BAP in the various types of manufacturing systems. Due to space limitations, only a few review papers restricted to the area of BAP in serial production lines are given in this Section. Even in this narrow area of research, there is a large number of publications. Many relevant references are not included for the sake of brevity. As stated in Papadopoulos et al. (2009), the solution of the BAP requires the use of two types of techniques: an evaluative technique to calculate throughput or any other performance measure such as the average work-in-process (WIP), and a generative or optimization technique to find the optimal or near optimal solution, i.e., the vector of buffer sizes which maximizes a given objective function. The latter may be a throughput or profit maximization or minimization of the number of buffer slots, or minimization of the average WIP in order to achieve a certain throughput level. A combination of more than one criterion may be also considered in the objective function (see Andijani and Anwarul, 1997). Evaluative techniques include exact analytical methods such as the Markov state model method, the stochastic automata network formalism and other Markovian structured methods, exact numerical methods, decomposition, aggregation /disaggregation, simulation, phase-type approach, holding time method, generalized expansion method, and other approximate methods. An excellent detailed overview of models of manufacturing flow line systems is given in Dallery and Gershwin (1992). An earlier review and comparison of models of automatic transfer lines was given by Buzacott and Hanifin (1978). Papadopoulos and Heavey (1996) provided a classification of models for production and transfer lines. Li et al. (2009) provided a comprehensive presentation of recent studies in the area of evaluative techniques of manufacturing systems. The criterion for choosing an evaluative technique is the speed of its convergence and the accuracy of the results. For short lines, exact numerical approaches are used (Hillier and Boling, 1967, and Papadopoulos, Heavey and O’Kelly, 1989 & 1990). For longer lines, decomposition (the pioneering work by Gershwin, 1987 and Dallery, David and Xie, 1988, Gershwin, 1994, Helber, 1999, Diamantidis et al., 2006), and aggregation techniques (the pioneering work by De Koster, 1987 and Lim, Meerkov and Top, 1990, Li and Meerkov, 2009) are more appropriate. Generative or optimization techniques include search algorithms such as complete enumeration and exact analytical methods (generally confined to short lines), the gradient Preprints of the 2013 IFAC Conference on Manufacturing Modelling, Management, and Control, Saint Petersburg State University and Saint Petersburg National Research University of Information Technologies, Mechanics, and Optics, Saint Petersburg, Russia, June 19-21, 2013 FrB2.2 1648