10381067 . solution methodology based genetic algorithm for multi-objective facility layout problem

Upload: shailesh-khare

Post on 03-Apr-2018

218 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    1/143

    Zagazig University

    Faculty of Engineering

    Solution Methodology Based Genetic

    Algorithm For Multi-Objective FacilityLayout Problem

    A Thesis submitted in fulfillment of the requirements for the degree of Master of

    Philosophy to Industrial Engineering and Systems Department

    By

    Raafat Hussien El-ShaerIndustrial and Systems Engineering Department, Faculty of Engineering,

    Zagazig University.

    Supervisors:

    Prof. Dr. Gamal Mohamed NawaraIndustrial and Systems Engineering Department, Faculty of Engineering,

    Zagazig University.

    Prof. Dr. Mohamed Abbas ShoumanOperations Research and Decision Support Department, Faculty of Computers

    and Informatics, Zagazig University.

    Dr. Hisham Mohamed El awadyIndustrial and systems Engineering Department, Faculty of Engineering,

    Zagazig University.

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    2/143

    Approval Sheet:

    Solution Methodology Based Genetic Algorithm ForMulti-Objective Facility Layout Problem

    A Thesis

    By

    Raafat Hussien El-Shaer

    Submitted in fulfillment of the requirements for the degree of Master of

    Philosophy to Industrial Engineering and Systems Department

    Approved as to style and content by:

    Prof. Dr. Faten Faheem Mahmoud

    Industrial and Systems Engineering Department, Faculty ofEngineering, Zagazig University.

    Prof. Dr. Mohamed Nashaat ForseIndustrial Engineering Department, Faculty of Engineering,

    Alexandria University.

    Prof. Dr. Gamal Mohamed NawaraIndustrial and Systems Engineering Department, Faculty of

    Engineering, Zagazig University.

    Prof. Dr. Mohamed Abbas ShoumanOperations Research and Decision Support Department,

    Faculty of Computers and Informatics, Zagazig University.

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    3/143

    I

    Acknowledgements

    No words can express my limitless gratitude and most sincere

    thanks to Allah against what He has given us. Special thanks

    and great appreciation are to Prof. Gamal Mohamed Nawara

    for his great tolerance and strong support. Special thanks and

    great appreciation are Also to Prof. Mohamed Abbas Shouman

    for his keen supervision and to the precious and valuable time he

    gave me. Thanks to his insightful comments, suggestions and

    valuable feedback he kindly offered, which helped me so much to

    accomplish my thesis.

    Also special thanks are due to Dr. Hisham Mohamed El awady

    for his great tolerance and sincere help, which guided me to

    fulfill this work.Moreover, my thanks are also due to many colleagues for various

    help .

    Finally, I would like to say thank-you to my family and my

    wife who encourage me and give me a quiet times to achieve

    this thesis.

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    4/143

    II

    Abstract

    The facility layout problem (FLP) concerns finding the most effectivearrangement of facilities, personals, and any resources required so as to

    minimize the costs associated with projected interaction among them. These

    costs may reflect material handling costs or preferences regarding

    adjacencies among departments. Because its one of the truly difficult ill-

    structured, multi-criteria, multi-objectives, and NP-completeness problem; the

    Artificial Intelligence techniques are one of the most powerful tools to deal

    with this problem category. Therefore, the prime objective of this study is to

    apply the Genetic Algorithm, (GA) as an efficient AI tool, capable of dealing

    with this problem and generating reasonable good solutions within a suitable

    CPU time.

    In order to achieve this objective, the following points have been fulfilled:

    1. A comprehensive review of literature on facilities layout is presented to

    show the progressive changes that have occurred over the past

    decades in techniques, the concepts and models. Especially, in Multi-

    Objectives facility layout problem (MOFLP).

    2. A brief overview of GAs is presented, which shows the motivated idea

    of the algorithm, the relevant operators, the encoding ways, the

    selection methods, crossover methods and mutation methods.

    3. GA is proposed for the problem under consideration, with

    comprehensive illustration for its building operators and parameters are

    discussed in details.

    4. GAMOFLP Software, written with C++ language, is developed for the

    proposed GA in point (3).

    This work addresses the second design stage of the cellular layout (i.e. inter-

    cell layout problem), which is a phase of the cellular manufacturing design,

    and is an important step before the design of finalized layout. The inter-cell

    layout problem seeks the best arrangement of cells based on their inter-

    relationships within the available area. In The proposed GA, a special

    permutation encoding is used to represent the chromosome and the

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    5/143

    III

    generation-based reproduction system with Elitism is applied using the

    tournament selection method, two-points crossover method and order change

    mutation method.

    In this work, the problem planning time is short and one floor is considered,

    and the available floor space can have restricted areas. The problem is

    affected by multiple criteria and has multiple objectives. The scaling problem

    among objectives is solved by a reasonable quadratic model.

    A solution methodology based GA for multi-objectives FLP (GAMOFLP) is

    developed to handle the conflicting criteria and normalize the objectives

    variables that affect the problem in order to get fair and reasonable good

    solution.

    The proposed GAMOFLP program is designed to tackle the following types of

    FLPs:

    1. Establish a layout for new facilities.

    2. Establish a layout for new facilities subject to restricted areas.

    3. Select the best position to add new facility (facilities) to an existing

    layout (plant).

    The validation of the proposed GAMOFLP system is verified using solved 20

    problems that have published in previous researches. The results have

    proved the efficiency and effectiveness of the proposed system and its

    capability of getting acceptable suboptimal solutions within convenient time. In

    details, related to the final solutions, the experiments show that the proposed

    GAMOFLP capable of finding better solutions than that those published or at

    least equal. Also comparatively with the required time to find the final

    solutions, in spite of its not fair tool for comparing, the results show the

    superiority of the system in providing better solutions within less CPU time

    than have published or at most in its limits.

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    6/143

    IV

    Contents

    Page

    Acknowledgement I

    Abstract II

    List of Tables VII

    List of Figures VIII

    List of Abbreviations X

    Chapter 1

    Introduction to Facility Layout Problem and Genetic Algorithms 1

    1.1 Facility Layout Problem 1

    1.2 Genetic Algorithms (GAs) 3

    1.3 Thesis Outline 4

    Chapter 2

    A Review and survey of FLPs 6

    2.1 Introduction 6

    2.2 FLP Formulation 6

    2.3 Computational Complexity 7

    2.4 Optimal Algorithms and Suboptimal Algorithms 9

    2.4.1 Optimal Algorithms 9

    2.4.2 Suboptimal Algorithms 9

    2.4.2.1 Constructive approaches 10

    2.4.2.2 Improving approaches 11

    2.4.2.3 Hybrid approaches 11

    2.4.2.4 Graphical approaches 11

    2.5 Intelligent Techniques for FLP 12

    2.5.1 Expert Systems and MCFLP 12

    2.5.2 Fuzzy Aystems and MCFLP 13

    2.5.3 Genetic Algorithms for MCFLP 14

    2.5.4 Intelligent Hybrid System for MCFLP 14

    2.6 General Comments 15

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    7/143

    V

    Chapter 3 Page

    Facility Layout Problem Formulation 16

    3.1 Introduction 16

    3.2 Quantitative Approach 17

    3.3 Qualitative Approach 18

    3.4 Multi-Objectives Approaches 19

    Chapter 4

    A Review of Genetic Algorithms 24

    4.1 What is a Genetic Algorithm? 24

    4.2 GA Parameters and Performance 25

    4.2.1 Crossover and Mutation 25

    4.2.2 Population Size 26

    4.2.3 Selection Methods 26

    4.2.4 Reproduction Methods 27

    4.3 On-line, Off-line and Best Individual Performance 28

    4.4 GA Application 29

    4.4.1 Layout Problems with GAs 29

    4.4.2 FLPs with GAs 30

    4.5 The Need for Good Solver Techniques 31

    Chapter 5

    Proposed Genetic Algorithm 32

    5.1 Why Genetic Algorithm? 33

    5.2 The Proposed Genetic Algorithm 33

    5.2.1 Chromosome Representation 34

    5.2.1.1 Representation of special layout configuration 34

    5.2.2 Initial Population 35

    5.2.3 Selection Method 35

    5.2.4 Crossover Operator 36

    5.2.5 Mutation Operator 37

    5.2.6 Reproduction System 37

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    8/143

    VI

    Page

    5.2.7 Evaluation Function 37

    5.2.8 Termination Criteria 37

    5.2.9 Algorithm Steps 38

    5.2.10 The default Values of The Proposed GA Parameters 39

    5.3 User Interface 41

    5.3.1 GAMOFLP Features 41

    5.3.2 GAMOFLP Screens 42

    Chapter 6

    System Evaluation and Validation 51

    6.1 Performance Evaluation 51

    6.2 Harraz's Procedure Comparison 51

    6.3 Comparison with Chen and Shas Procedure 60

    6.4 Common Test Problems 62

    Chapter 7Conclusions and Future Work 65

    7.1 Snapshot About Thesis 65

    7.2 Conclusion 65

    7.3 Scope and Recommendation of Future Work 66

    References 68

    Appendix ( A ): Hill-Climbing Investigation Method 76Appendix ( B ): Test Problems Data 79

    Appendix ( C ): Proposed GAMOFLP Code 84

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    9/143

    VII

    List of Tables

    Page

    Table 2.1 : Classification of FLP solvers. 8

    Table 2.2 : Constructive approaches. 10

    Table 5.1 : GA parameters to be investigated. 40

    Table 5.2 : Hill-climbing investigation procedure. 40

    Table 6.1 : Best solution qualities for test problem in Harraz [64]. 52

    Table 6.2 : Best solution qualities for test problem in Harraz [64]. 52

    Table 6.3 : Best solution qualities for test problem in Harraz [64]. 54

    Table 6.4 : Best solution qualities for test problem in Harraz [64]. 55

    Table 6.5 : Best solution qualities for test problem in Harraz [64]. 56Table 6.6 : Best solution qualities for test problem in Harraz [64]. 57

    Table 6.7 : Best solution qualities for test problem in Harraz [64]. 58

    Table 6.8 : Best solution qualities for test problem in Harraz [64]. 59

    Table 6.9 : Best solution qualities for test problems in Chen and Sha [21]. 60

    Table 6.10: Best solution qualities for test problem in Chen and Sha [21]. 61

    Table 6.11: Best solution qualities for the eight test problems in Nugent et al.[41]. 63

    Table 6.12: Average solution qualities for test problems in Nugent et al. [41]. 64

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    10/143

    VIII

    List of Figures

    Page

    Figure 3.1: Two algorithms performance. 28

    Figure 3.2 : Chromosome representation for an identical FLP. 30

    Figure 5.1 : Phenotype representation layout. 34

    Figure 5.2 : The available layout space with restricted area. 35

    Figure 5.3 : Available layout space of restricted area and pre-specified facilities. 35

    Figure 5.4 : PMX crossover method. 37

    Figure 5.5 : Genetic algorithm flow chart. 39

    Figure 5.6 : Some Basic restricted shapes for the FLPs available space. 42

    Figure 5.7 : Available space and pre specified facilities. 42

    Figure 5.8 : Screen no. ( 1 ) Genetic Algorithm parameters. 43

    Figure 5.9 : Screen no. ( 2 ) Criteria type. 43

    Figure 5.10: Screen no. ( 3 ) Layout space. 44

    Figure 5.11: Screen no. ( 4 ) Available layout space. 45

    Figure 5.12: Screen no. ( 5 ) Pre-specified facilities. 45

    Figure 5.13: Screen no. ( 6 ) Entering relationship matrix. 46

    Figure 5.14: Screen no. ( 7 ) Entering flow matrix. 46Figure 5.15: Screen no. ( 8 ) Entering cost matrix. 46

    Figure 5.16: Screen no. ( 9 ) Entering the objectives weight. 47

    Figure 5.17: Screen no. (10) Confirming relationship matrix data. 47

    Figure 5.18: Screen no. (11) Adjusting coding system values. 48

    Figure 5.19: Screen no. (12) confirming flow matrix data. 48

    Figure 5.20: Screen no. (13) Solution progress. 49

    Figure 5.21: Screen no. (14) Final solution and improvement path. 49

    Figure 5.22: Best illustrative example solution after 500 generation. 50

    Figure 6.1: Five-department problem % improvement in the OFV and

    CPU computation time. 52

    Figure 6.2: Six-department problem% improvement in the OFV and

    CPU computation time. 53

    Figure 6.3: Seven department problem % improvement in the OFV and

    CPU computation time. 54

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    11/143

    IX

    Page

    Figure 6.4: Eight department problem% improvement in the OFV and

    CPU computation time. 55

    Figure 6.5: Nine department problem % improvement in the OFV and

    CPU computation time. 56

    Figure 6.6: Ten department problem % improvement in the OFV and

    CPU computation time. 57

    Figure 6.7: Eleven department problem % improvement in the OFV and

    CPU computation time. 58

    Figure 6.8: Twelve department problem % improvement in the OFV and

    CPU computation time. 59

    Figure 6.9: 20-department problem % improvement in the OFV. 62

    Figure A.1 : Flow, adjacency and distance data for a randomly generated

    25-cell problem 76

    Figure B.1 : Flow, adjacency and distance data for a 5-cell problem. 79

    Figure B.2 : Flow, adjacency and distance data for a 6-cell problem. 79

    Figure B.3 : Flow, adjacency and distance data for a 7-cell problem. 79

    Figure B.4 : Flow, adjacency and distance data for a 8-cell problem. 79

    Figure B.5 : Flow, adjacency and distance data for a 9-cell problem. 80

    Figure B.6 : Flow, adjacency and distance data for a 10-cell problem. 80

    Figure B.7 : Flow, adjacency and distance data for a 11-cell problem. 80

    Figure B.8 : Flow, adjacency and distance data for a 12-cell problem. 81

    Figure B.9 : Flow, adjacency and distance data for an 8-cell problem. 81

    Figure B.10: Flow, adjacency and distance data for a 12-cell problem. 81

    Figure B.11: Flow, adjacency and distance data for a 15-cell problem. 82

    Figure B.12: Flow, adjacency and distance data for a 20-cell problem. 83

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    12/143

    X

    List of Abbreviations

    AI Artificial Intelligence

    AHP Analytic Hierarchy Process

    ALDEP Automated Layout Design ProgramANNs Artificial Neural Networks

    AS Ant System

    C++ C++ Language

    CM Cellular Manufacturing

    CORELAP Computerized Relationship Layout Planning

    CRAFT Computerized Relative Allocation of Facilities Technique

    DEA Data Envelopment Analysis

    DM Decision-Maker

    ES Expert Systems

    FATE A new Construction Algorithm for Facilities Layout

    FDMS Fuzzy Decision-Making SystemsFL Fuzzy Logic

    FLEXPERT Fuzzy-integrated Expert System for Facility Layout

    FLP Facility Layout Problem

    GA Genetic Algorithm

    HC Hill-Climbing Method

    HSs Hybrid Systems

    KBDSS Knowledge-Based Decision

    LB Lower Bound

    LPA Linear Placement Algorithm

    MAT Modular Allocation Technique

    Max.Iter Maximum Iteration NumberMCDM Multi-Criteria Decision-Making

    MCFLP Multi-Criteria Facility Layout Problem

    MOFLP Multi-Objectives Facility Layout Problem

    N Total Number of Facilities

    NP Non-Deterministic Polynomial

    OFV Objective Function Value

    OV Objective Value

    Pc Crossover Probability

    PLANET Plant Layout Analysis and Evaluation Technique

    Pm Mutation Probability

    PMX Partially Matched Crossover

    PopSize Population Size

    QAP Quadratic Assignment Problem

    QSP Quadratic Set-covering Problem

    SA Simulated Annealing

    SBL Shape-based Block Layout

    TFC Total Flow Cost

    TNR Total Numerical Rating

    Ts Tournament Size

    VLSI Very Large Scale Integrated

    WIP Work-in-process

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    13/143

    Chapter 1

    Introduction to: Facility Layout Problem

    and Genetic Algorithms

    1.1 Facility Layout Problemhe facility layout problem, FLP, deals with finding the most effective

    physical arrangement of facilities, personal, and any resources required

    to facilitate the production of goods and services. It has attracted the attention

    of many researchers due to its practical utility and interdisciplinary importance.

    The production function of a manufacturing company is significantly affected

    by the layout of its manufacturing shop. While a well designed layout can

    considerably improve the efficiency of the shop, a poor one leads to increased

    work-in-process (WIP) overloading the material handling system and

    contribute to inefficient set-ups, longer queues, etc. Ham et al. [1]. Material

    handling and layout related costs have been estimated to be about 20%-50%

    of the operating expenses in manufacturing systems, Tompkins and White [2].Historically, two basic approaches have most commonly been used to

    generate desirable layouts: a qualitative one and a quantitative one. These

    approaches are usually used one at a time when solving a facility layout

    problem.

    With qualitative approaches, layout designers provide subjective evaluations

    of desired closeness between departments. Then, overall subjective

    closeness ratings between various departments are maximized. Thesesubjective closeness ratings are: A (absolutely necessary), E (essentially

    T

    1

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    14/143

    Chapter 1 : Introduction to FLP & GA 2

    important), I (important), O (ordinary), U (unimportant) and X (undesirable), to

    indicate the respective degrees of necessity that two given departments be

    located close together. Layout designers may then assign numerical values to

    the ratings such that they have the ranking A > E > I > O > U> X. Seehof and

    Evans [3], Lee and More [4], Muther and McPherson [5] and Muther [6] have

    developed algorithms based on qualitative criteria to obtain final layouts.

    These different qualitative approaches are distinguished primarily by the

    scoring methods used for the closeness ratings. In this aspect, the numerical

    values used by Sule [7] and Harmonosky and Tothero [8] for ratings are A= 4,

    E= 3, I= 2, O= 1, U= 0 and X= -1. Also, the ALDEP procedure presented by

    Seehof and Evans [3] used the numerical values: A= 64, E= 16, I= 4, O= 1,

    U= 0 and X= -1024.

    Quantitative approaches involve primarily the minimization of material

    handling costs between various departments. The quadratic assignment

    problem (QAP) formulation for assigning N facilities to N mutually exclusive

    locations is the most typical model used. Gilmore [9], Lawler [10] and, Gavett

    and Plyter [11] have offered exact solution procedures using branch-and-

    bound techniques. However, The QAP formulation belongs to the class of NP-

    complete problems Garey and J ohnson [12], and no definite method can

    achieve optimal solution in a reasonable time when 15 or more facilities are

    considered. Consequently, many heuristic algorithms have been developed

    for achieving a trade-off between computation time and the efficiency of the

    final solution Kusiak and Heragu [13]. The proposed approach in this work is

    also a heuristic one.

    Many researchers are seeking about the appropriateness of a single criterion

    objective selection in solving the facility layout problem due to the drawbacks

    of using qualitative and quantitative approaches separately. The major

    limitations on quantitative approaches are that they consider only relationships

    that can be quantified and no qualitative factors are considered. The

    shortcoming of qualitative approaches is their strong assumption that all

    qualitative factors can be aggregated into one criterion. In real life, the facility

    layout problem must consider quantitative and qualitative criteria and this falls

    into the category of the multi-objective facility layout (MOFL) problem.

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    15/143

    Chapter 1 : Introduction to FLP & GA 3

    The primary purpose in solving the MOFL problem is to generate efficient

    alternatives that can be presented to the decision maker for his or her

    selection. Malakooti and Tsurushima [14] classified three methods for solving

    the MOFL problem:

    a) Generate a set of efficient layout alternatives and then present it to the

    decision maker;

    b) Assess the decision makers preferences first, and then generate the

    best layout alternative; and,

    c) Use an interactive method to find the best layout alternative.

    The proposed approach in this work falls into the category of the type (a)

    methods in terms of generating good-quality solutions using an effective

    heuristic algorithm. As a matter of fact, Rosenblatt [15], Dutta and Sahu[16],

    Fortenberry and Cox [17], Waghodekar and Sahu [18], Urban [19] and

    Houshyar [20], Harmonosky and Tothero [8] all developed QAP formulations

    by specifying different objective weights to generate the best layout [see

    section (3.4)]. However, there are two inadequacies in these approaches:

    1. All factors are not represented on the same scale.

    2. Measurement units used for objectives are incomparable.

    Chen and Sha [21] developed a heuristic approach to overcome the above-

    mentioned inadequacies by reasonably normalizing all objectives of MOFLP,

    and handling qualitative and quantitative information in a similar fashion. In

    the same direction, in this work, an approach based genetic algorithm using

    Chen and Shas model to deal with this problem and comparative analysis

    have been presented.

    1.2 Genetic Algorithms (GAs)Genetic algorithms attempt to mimic the biological evolution process for

    discovering good solutions. They are based on a direct analogy to Darwinian

    natural selection and mutations in biological reproduction and belong to a

    category of heuristics known as randomized heuristics. They employ

    randomized choice operators in their search strategy and do not depend on

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    16/143

    Chapter 1 : Introduction to FLP & GA 4

    complete a priori knowledge of the features of the domain. These operators

    have been conceived through abstractions of natural genetic mechanisms

    such as crossover and mutation and have been cast into algorithmic forms

    Arunkumar and Chockalingam [22]. Repetitive executions of these heuristics

    need not yield the same solution.

    A genetic algorithm maintains a collection or population of solutions

    throughout the search. It initializes the population with a pool of potential

    solutions to the problem and seeks to produce better solutions by combining

    the better of the existing ones through the use of one or more genetic

    operators. solutions are chosen at each iteration with a bias towards those

    with the best objective values. With various mapping techniques and an

    appropriate measure of fitness of solutions (i.e., objective function value), a

    genetic algorithm can be tailored to evolve a solution for many types of

    problems, including optimization of a function or determination of the proper

    order of a sequence.

    Theoretical analyses suggest that genetic algorithms can quickly locate

    high performance regions in extremely large and complex search spaces. In

    addition, the distributed and repeated sampling can lead to some natural

    insensitivity to noisy feedback. The genetic algorithm based heuristics are

    highly suited for application to large instances of problems that are hard to

    model and for which no satisfactory tailored algorithms are available. There

    are many variations and refinements, but basically, any genetic algorithm

    contains reproduction, crossovers and mutation.

    GAs have been used to find good solution to various NP-complete

    problems such as time-tabling problems Ross et al [23] and cable routing

    problems Kloske and Smith [24]; and have shown good performance in many

    applications. Actually some papers such as Cohoon et al [25] and Tam [26]

    suggest GAs superiority in FLPs. Therefore, GAs are a promising approach to

    FLPs.

    1.3 Thesis OutlineThe scope of the work is not to solve a planning problem, it is to handle the

    multi-criteria aspect that affects the idealized relative location of cells within a

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    17/143

    Chapter 1 : Introduction to FLP & GA 5

    floor plan. Due to the combinatorial and ill-structured nature of the problem, a

    solution methodology based GA heuristic is developed to suggest an

    acceptable solution. In the proposed GA, a special chromosome

    representation is applied to represent some constraints such as restricted

    locations and pre-specified facilities. Moreover, an investigation using Hill-

    climbing method is applied to tune up the parameters of the proposed GA.

    The approach is validated by many recommended test problems taken from

    previous papers and compared with other multi-objective approaches.

    This thesis consists of seven chapters including this chapter. In Chapter 2,

    FLPs are reviewed and surveyed especially, multi criteria. The Genetic

    Algorithms and their features are reviewed in Chapter 3. In Chapter 4, the FLP

    formulation is reviewed and proposed formulation is mentioned. The

    implementation details of the proposed GA are described in Chapter 5.

    Chapter 6 presents comparative study between the proposed approach and

    other MOFLP approaches. Finally, in Chapter 7, the conclusions of this

    research work and some recommendations for future work are highlighted.

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    18/143

    Chapter 2

    A Review and survey of FLPs

    2.1 Introduction

    any sorts of layout can be mentioned. For example, bin-packing

    problems such as Smith [27] and Falkenauer [28] try to maximize the

    storage packets number; VLSI chip layout problems like Larmore et al [29]

    and Saab and Rao [30] aim to minimize the area occupied by chips. In

    particular, facility layout problems (FLPs) may be one of the largest fields.

    This is because there is usually a problem to minimize total traffic cost

    between facilities in a building. For example, FLP can be applied to

    manufacturing machines in a factory Souilah [31], to people traffic in an office

    Steadman [32], and to backboard wiring on an electrical board Steinberg [33].

    In FLPs, under given conditions including facility size and flow between each

    pair of them, the problem is to minimize the total flow cost.

    2.2 FLP Formulation

    Though some other models have been suggested as shown in Kusiak and

    Heragu [13], the quadratic assignment problem (QAP) described in Koopmans

    and beckmann [34] may be regarded as the basic representation of FLPs. The

    M

    2

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    19/143

    Chapter 2 : A Review and survey of FLPs 7

    name of (QAP) was so given because the objective function is a second-

    degree function of variables and the constraints are linear functions of

    variables. In spite of the QAP has been frequently used to model the FLP, this

    does not mean that all FLPs can be formulated as a QAP. In the case of

    machine layout problem (MLP) in which the locations of machines are not

    known initially, QAP cannot be formulated.

    The second formulation for FLP is a quadratic set-covering problem (QSP)

    Bazaraa [35]. In this formulation, the total area occupied by all facilities is

    divided into a number of blocks where each facility is assigned to exactly one

    location and each block is occupied by at most one facility. A disadvantage of

    QSP formulation is that the problem size increases as the total area occupied

    by all the facilities is divided into smaller blocks Bazaraa [35].

    The FLP is formulated using several integer-programing formulations. Lawler

    [10] was the first to formulate the facility layout problem as a linear integer-

    programing problem; Love and Wong [36] proposed a simple integer

    programming formulation to the QAP. Kaufman and Broeckx [37] developed a

    linear mixed integer program to formulate the FLP and Ritzman et al [38]

    formulated a large mixed integer goal programming model for assigning

    offices in buildings.

    Flouds and Robinson [39] developed a Graph-theoretic approach model. in

    this approach, it is assumed that the desirability of locating each pair of

    facilities adjacent to each other is known. In this model, a closeness rating

    indicating desirability of locating facility i adjacent to facility j is assumed. The

    model seeks to maximize the closeness rating of the facilities.

    In addition of the above formulations, Rosenblatt [15] developed a model,

    which minimize the flow cost of material and maximize a closeness rating

    measure (multi-objective FLP). These two objectives are conflicting

    objectives. Rosenblatt [15], Dutta and Sahu [16] developed heuristic

    algorithms to solve the model.

    2.3 Computational Complexity

    Because of there are N! ways of putting N facilities into N locations the QAPs

    objective function can take N! different values at most. As a result, in order to

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    20/143

    Chapter 2 : A Review and survey of FLPs 8

    get the best layout, all the N! patterns should be estimated. Nevertheless,

    since N! grows extremely large ifNgoes large, it is impossible to search for all

    patterns in polynomial time. So, Sahni and Gonzalez [40] showed that the FLP

    consisting of many facilities is a NP-complete problem. To find an optimal

    solution to the fifteen-facility problem in Nugent et al [41], more than an hour

    of CPU time was required on a CDC CYBER 76. Among the eight test

    problems in Nugent et al [41], the largest problem for which an optimal

    solution was found was the fifteen-facility layout problem. This is considered

    as a major constraint.

    Between the late 1950s and the mid 1990's, a number of algorithms have

    been developed to solve the FLP. These algorithms may be classified as:

    Optimal algorithms,

    Suboptimal algorithms.

    Table 2.1 lists FLP solvers as optimal and suboptimal procedures.

    Table 2.1: Classification of FLP solvers Kazuhiro [42].

    Procedure Description

    Optimal algorithms

    Looking for the best layout (How to prune the

    redundant alternatives is important.) [43],[11].

    Suboptimal algorithms Looking for reasonably good layout

    Constructive approach Putting one facility to another

    (Usually facilities having heavier load are

    considered prior to lighter ones) [44],[45].

    Improving approach From a random layout, exchanging some

    selected facilities (selection strategy is

    important.) [46],[47],[41]

    Hybrid approach A combination of constructive and improving

    approach. [48],[49],[50]

    Graphical approach Analytical approach (a sort of constructive

    approach) [51],[52]

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    21/143

    Chapter 2 : A Review and survey of FLPs 9

    2.4 Optimal Algorithms and Suboptimal Algorithms

    The next subsections present a review about the two procedures.

    2.4.1 Optimal Algorithms

    Optimal algorithms search for the best solution by some heuristics like branch

    and bound Land [43], Bazaraa and Elshafei [53] or Gavett and Plyter [11] and

    cutting plane Bazaraa and Sherali [54], whereas suboptimal procedures look

    for reasonably good solutions by various strategies. However, owing to the

    NP-completeness, the approaches for optimal solution may be impractical

    especially when number of facilities increases Hanan and Kurtzberg[55]. The

    computational results reported in Lavalle and Roucairol [56] indicate that the

    parallel branch and bound algorithm requires high computation time for layout

    problems with twelve or more facilities. Moreover, the largest facility layout

    problems are solved optimally by a cutting plane algorithm is a layout problem

    with eight facilities.

    A common experience of optimal algorithms is that optimal solution is found

    early in the branching process but is not verified until a substantially high

    number of solutions have been enumerated Burkard and Stratmann [48],

    Bazaraa and Kirca [49]. This prompted researchers to terminate the branch

    and bound process prematurely without verifying optimality. This resulted in

    heuristic branch and bound algorithms.

    Burkard [57] listed two criteria for premature termination of branch and bound

    process; termination based on time limits and termination based on upper

    bounds quality.

    2.4.2 Suboptimal algorithms

    The optimal algorithms discussed above have the following disadvantages [13]:

    1. Computational time requirement is high.

    2. The largest problem size solved optimally is a problem of fifteen facilities.

    This directs the researchers to concentrate on suboptimal algorithms for

    solving the FLP. According to Kusiak and Heragu [13], the suboptimal

    algorithms can be categorized into four approaches as exhibited in Table 2.1.

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    22/143

    Chapter 2 : A Review and survey of FLPs 10

    2.4.2.1 Constructive approaches

    Some algorithms like MAT in Edwards et al [44] and LPA in Neghabat

    [45] are called constructive approaches, since they make physical layouts by

    adding one facility to another. In these approaches, deciding the order of

    putting facilities is generally important. In this aspect, LPA begins by placing

    the two facilities of the highest flow among each other; then another facility is

    placed, which has the highest flow with the located facilities, as close as

    possible to them; and so on. MAT ranks pairs of facilities according to their

    flow values and location pairs according to their distance values and uses this

    information to determine the layout. Also there are many constructive

    approaches which use Rel-chart to determine the ordering of facilities to

    construct a layout like ALDEP in Seehof and Evans [3], CORLAP in Lee and

    Moore [4], and RMA Comp I in Muther and McPherson [5]. Table 2.2 lists

    constructive approaches summary.

    Table 2.2 : Construct ive approaches.

    Use the following data to adjust

    facilities building orderConstruction

    algorithmFrom-to-chart Rel-chart

    Year,Ref.

    HC66 * 1966,[47]

    ALDEP * 1967,[3]

    CORLAP * 1967,[4]

    RMA Comp I * 1970,[5]

    MAT * 1970,[44]

    PLANET * 1972,[58]

    LSP * 1972,[59]

    LPA * 1974,[45]

    FATE * * 1978 [60]

    INLAYT * 1980,[61]

    FLAT * 1986,[62]

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    23/143

    Chapter 2 : A Review and survey of FLPs 11

    2.4.2.2 Improving approaches

    In improving approaches, there is always an initial solution, which is

    often randomly generated. Based on this initial solution, systematic

    exchanges between facilities are made in order to get better solutions. The

    exchanges continue until either a reasonably good solution is obtained or a

    certain allowed time passes. Hence, the solution quality of these approaches

    depends upon the initial layout evaluation, and the selection strategy of

    exchanging facilities is critical. In this aspect, while CRAFT in Buffa et al [46]

    has no limitation of facilities exchange, HC63-66 in Hiller and Connors [47]

    restricte the facilities exchange in the aim of computation time reduction. Also,

    some algorithms like Biased Sampling Method in Nugent et al [41] select

    facilities stochastically to save time.

    2.4.2.3 Hybrid approaches

    hybrid approaches are used to obtain the merits of both approaches

    above. This is, starting from a layout created by a constructive algorithm; the

    layout is modified by some improving algorithms. For instance, Burkard and

    Stratmann [48] create an initial layout by branch and bound technique under

    time limit; then, improve it by exchanging two or three facilities at one time.

    2.4.2.4 Graphical approaches

    Graphical approachesput importance on facilities adjacencies. In these

    approaches, each facility and each adjacency of two facilities are represented

    by a node and an arc, respectively. Therefore, a feasible layout must be a

    planar graph, which can be drawn on a plane without any intersections of

    arcs. Thus, as Kusiak and Heragu [13] suggested, these approaches

    classified as sorts of construction approaches because they construct a larger

    graph by adding a new node to a planar graph so that the new graph can be

    still planar and that its flow cost can be as small as possible. However,

    according to Foulds [51], these approaches may be able to solve FLPs with at

    most fifteen facilities in reasonable computation time.

    The major drawbacks of the aforementioned approaches lie in the fact that thesearch for the best layout is not very efficient, is not easily obtainable, and the

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    24/143

    Chapter 2 : A Review and survey of FLPs 12

    multi-objective nature are not considered in the problem Hiller and Connors

    [47]. As a matter of fact, facility layout problem can be considered one of the

    truly difficult ill-structured, multi-criteria decision making and combinatorial

    optimization problem. According to Malakooti and Tsurushima [14] the layout

    problem is an ill-structured problem because:

    There are multiple criteria that must be considered to test alternate

    layouts.

    It is hard to determine a problem space that can represent all

    characteristic of the problem.

    There are many domain-specific and problem-specific constraints.

    Due to the combinatorial nature of facility layout problem, finding an optimal

    solution is very complex and tedious task, especially if the number of facilities

    is large. Still finding out for new and recent developments rather than

    conventional approaches to overcome the aforementioned drawbacks.

    Intelligent techniques such as Expert systems (ESs), Fuzzy logic (FLs),

    Genetic algorithms (GAs), Ant systems (ASs) , Artificial neural networks

    (ANNs) and Aybrid systems (HSs) have been used as new advancements for

    the tackled problem.

    2.5 Intelligent Techniques for FLP

    Intelligent techniques were introduced to the field of facilities layout in

    the early 1980s. Most of these systems are briefly reviewed taking into

    account the multi-criteria concept (MCFLP).

    2.5.1 Expert Systems and MCFLP

    Kumara et al [63] have developed a heuristic-based ES. They have

    defined the facilities layout problem as a multi-objective problem and have

    outline a methodology to handle the qualitative constraints in conjunction with

    heuristic procedures for quantitative parameters.

    Malakooti and Tsurushima [14] have developed an ES for multi-criteria facility

    layout. Their approach is based on expert systems and multi-criteria decision-

    making (MCDM). The expert system interacts with the decision-maker (DM),

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    25/143

    Chapter 2 : A Review and survey of FLPs 13

    and reflects the DMs preferences in the selection of rules and priorities. The

    approach consists of two parts: (1) construction of a layout based on a set of

    rules and restrictions, and (2) improvement of the layout based on interaction

    with DM. The MCDM expert system approach considers and incorporates the

    multiple-criteria in these two parts as follows: First, it uses priorities on the

    selection of rules, adjacency of departments, and departments for

    construction purposes. Second, it uses different objectives such as material

    handling cost, flexibility, and material-handling time for paired comparison of

    generated layouts for improvement purposes.

    Harraz [64] has developed a knowledge-based decision support system for

    multi-criteria facility layout. The system works in a tandem mode. It combines

    a rule-based module with an optimization module. The rule-based module

    enables the user to assign different priorities for criteria and generates a

    layout based on a set of rules. The resultant layout is seeded optionally to the

    improvement algorithm to find a better configuration for the solution. The

    improvement module is based on the simulated annealing (SA) global

    optimization algorithm. The proposed algorithm in the current work will use the

    test problems used in Harraz [64], as a comparison.

    2.5.2 Fuzzy Systems and MCFLP

    Raoot and Rakshit [65] have developed a linguistic pattern approach

    for multiple criteria facility layout problems. A multiple criteria model is

    formulated using the basic concept of linguistic pattern and a heuristic

    procedure is proposed to generate a set of efficient alternative layouts. The

    problem of selection of the facility layout from a set of alternatives, which

    satisfy different objectives and restrictions to known degrees, is considered as

    a MCDM problem and the ELECTRE method based on out ranking relations

    approach, is used to select the best layout. Dweiri and Meier [66] have

    established a vigorous methodology, based on fuzzy set theory, to improve

    the facility layout process. The AHP is used to find the weights of both

    qualitative and quantitative factors, which affects the closeness rating

    between departments in a plant. FUZZY, a computer program developed

    based on the fuzzy decision-making systems (FDMS), is used to generate the

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    26/143

    Chapter 2 : A Review and survey of FLPs 14

    activity relationship charts. These charts are used by FZYCRLP, a modified

    version of CORELAP, to develop the layouts. FELAP, another program based

    on FDMS, is used to evaluate the layouts. This evaluation method uses the

    distances and the relationships between departments to score the layout.

    2.5.3 Genetic Algorithms for MCFLP

    Hamamoto et al [67] developed a genetic algorithm based- facility

    layout method with an embedded simulation model for the pharmaceutical

    industry. This method allows the user to select the objectives that are

    important in each particular layout design in the pharmaceutical industry.

    2.5.4 Intelligent Hybrid Systems for MCFLP

    Badiru and Arif [68] have developed, FLEXPERT, a fuzzy-integrated

    expert system for facility layout. FLEXPERT considers the multi-criteria nature

    of the layout problem and the fuzziness of the input data through the

    integration of an expert system and a fuzzy algorithm with a commercial

    facility layout program (BLOCPLAN). The system generates the best layout

    that satisfies the qualitative as well as the quantitative constraints on the

    layout problem. Yang and Kuo [69] proposed a hierarchical analytic hierarchy

    process (AHP) and data envelopment analysis (DEA) approach to solve a

    plant layout design problem. A computer-aided layout-planning tool was used

    to generate a considerable numbers of layout alternatives as well as to

    generate quantitative decision-making unit (DMU) outputs. The qualitative

    performance measures were weighted by AHP. DEA was then used to solve

    the multiple-objective layout problem. Lee and Lee [70] presented a shape-

    based block layout (SBL) approach for solving facility layout problem. The

    SBL approach employs hybrid genetic algorithm to find good solution. The

    objective function of SBL approach minimizes total material handling cost and

    maximizes space utilization. Azadivar and Wang [71] presented a facility

    layout optimization technique, using simulation and genetic algorithms, that

    takes into consideration the dynamic characteristics and operational

    constraints of the system, and is able to solve the facility layout design

    problem based on a system's performance measures, such as the cycle time

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    27/143

    Chapter 2 : A Review and survey of FLPs 15

    and productivity. Genetic algorithms are used to optimize the layout for

    manufacturing effectiveness while simulation serves as a system performance

    evaluation tool.

    2.6 General Comments

    As a result from the survey, it is concluded that the facility layout

    problem still requires effective approach that overcomes the mentioned

    inadequacies. In this work, an effective approach using the Chen and Shas

    formulation is presented. The approach reasonably normalizes all objectives

    of the MOFL problem, and handling qualitative and quantitative and any other

    objectives information in similar fashion. Because of existing optimizationmethods are computationally inefficient when large numbers of facilities are

    involved, stochastic heuristic methods are more appropriate for generating

    effective layouts.

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    28/143

    Chapter 3

    Facility Layout Problem Formulation

    3.1 Introduction:

    he facility layout problem formulation considers the arrangement ofN

    facilities toMlocations (N

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    29/143

    Chapter 3: Facility Layout Problem Formulation 17

    ])()[( 22),( klklkl yyxxD

    Two approaches are exist for solving facility layout problems, quantitative

    approach and qualitative approach [72]. These approaches are:

    3.2 Quantitative Approach

    This approach tries to assign N facilities to M locations in order to

    minimize the material handling cost between departments. In this approach,

    FLP is formulated as Quadratic Assignment Problem (QAP) [34] because the

    objective function is a second-degree polynomial function of the variables, and

    the constraints are identical of the constraints of the assignments problems.

    The objective function is defined as:

    The value W(i,j) is the number of pallets trips required for transferring all the

    parts, which use the two facilities consecutively. Note that the quantity W(i,j)

    may also be weighted by any combination of parameters to consider special

    material handling requirements, part weights, actual material handling costs,

    etc. to quantify the distanceD(l,k) between the locations landk. Two traditional

    measures can be used:

    Cartesian distance; i.e.:

    otherwise.0and,locationtoassignedisfacilityif1ofnumberbinaryis

    and;andlocationbetweendistancetheis

    .facilityandfacilitybetweenflowmaterialofnumbertheis

    Where

    ________________

    ,10

    (3),...,2,11

    (2),...,2,11

    :_

    (1).

    ),(

    ),(

    ),(

    ),(

    1 ),(

    1 ),(

    1

    1 1 ),(),(),(),(11

    vuX

    klD

    jiW

    MN

    vuallfororX

    MvX

    NuX

    tosubject

    XXDWQMin

    vu

    kl

    ji

    vu

    N

    u vu

    M

    v vu

    N

    i

    N

    ij kjliklji

    M

    l

    M

    lk

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    30/143

    Chapter 3: Facility Layout Problem Formulation 18

    Or the rectilinear distance; i.e.:

    Where (xl,yl) and (xk,yk) are the coordinates of the geometric centers of

    locations whose facility i and facilityj respectively. Since the coordinate space

    considered is discrete and finite, each point in the space is assigned a unique

    position number in order to simplify the analysis. The problem, thus, consists

    of determining the position number corresponding to each ofN facilities, in

    order to minimize the objective function ofequation (1). Constraint (2) ensures

    that the facilities do not overlap when they are assigned to adjacent position.

    Furthermore, constraint (3) ensures each position cannot be occupied by

    more than one facility.

    3.3 Qualitative Approach

    It includes assigning N facilities to M locations in order to locate

    departments that utilize common materials or utilities adjacent to one another,while separating departments for reasons of safety, noise, heat, etc. The

    approach uses QAP formulation [34] and employs relationship chart (REL

    chart) as input to this process. The using of lettering system provides the

    desirability associated of each pair of departments located adjacently. The

    objective function is defined as:

    )()(),( klklkl yyxxD

    MN

    vuallfororX

    MvX

    NuX

    tosubject

    XXLRQMax

    vu

    N

    u vu

    M

    v vu

    N

    i

    N

    ij

    M

    l

    M

    lk kjliklji

    ,10

    )6(,...,2,11

    )5(,...,2,11

    :_

    )4(.

    ),(

    1 ),(

    1 ),(

    1

    1 1 1 ),(),(),(),(2

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    31/143

    Chapter 3: Facility Layout Problem Formulation 19

    3.4 Multi-Objectives Approaches

    Some techniques have been presented for solving the problem with

    both qualitative and quantitative objectives. These involved combining the two

    formulations into one single formulation and consider both quantitative and

    qualitative information.

    J acobs [73] and Shang [74] formulated the problem as multi-criteria and

    consider only one quantitative or qualitative factor in the objective function.

    The problem is solved using computer heuristics similar to those used for

    single goal problems.

    Other formulation allow problem to be solved using two quantitative and

    qualitative factors simultaneously such as those proposed by Harraz [64],

    Sarin et al [75], and Chen and Sha [21]. The following section will present an

    overview of MOFLP approaches:

    The QAP formulation of the MOFL problem is shown in equations (7) to (10):

    .locationatfacilityandlocationatfacilitylocatingofcostThe

    otherwise,

    ,locationtoasssignedisfacilityif

    0

    1

    Where

    (10),10

    (9),...,2,11

    (8),...,2,11

    :_

    )7(.

    ),(

    ),(

    1 ),(

    1 ),(

    1

    1 1 1 ),(),(

    kjliA

    vu

    X

    MN

    vuallfororX

    MvX

    NuX

    tosubject

    XXAQMin

    ijlk

    vu

    vu

    N

    u vu

    Mv vu

    N

    i

    N

    ij

    M

    l

    M

    lk kjliijlk

    otherwise.(0)and,locationtoassignedisfacilityif)1(esnumber takBinary

    otherwise

    adjacentarelocationstwowhen the

    0

    1

    siteandsitebetweeneightLocation w

    andfacilitybetweenvalueratingCloseness

    Where

    ),(

    ),(

    ),(

    vuX

    (k)(l)L

    jiR

    vu

    kl

    ji

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    32/143

    Chapter 3: Facility Layout Problem Formulation 20

    Aijlk in equation (7) is a cost variable representing the combination of

    quantitative and qualitative measures in MOFL models. Equation (8) ensures

    that each location contains only one facility. Equation (9) ensures that each

    facility is assigned to only one location. As mentioned Chen and Sha [21],

    these models are classified into four categories:

    (1) Rosenblatt [15] and Dutta and Sahu [16] defined the cost term as:

    Where Cijlk is the total material handling cost, Rijlk is the total

    closeness rating score, andFCand WRare weights assigned to the

    total material handling cost and to the total rating score.(2) Foretenberry and Cox [17] defined the cost term as:

    WhereFij is the workflow between two facilities, Dlk is the distance

    between two locations andRij is the closeness rating desirability of

    the two facilities.

    (3) Urban [19] defined the cost term as:

    Where Cis a constant weight that determines the importance of the

    closeness rating of workflow.

    (4) Khare et al [76] defined the cost term as:

    Where W1 and W2 are weights assigned to the work flow and to

    the closeness rating.

    The listed models are similar in nature, and vary only in stating the

    relationship between the cost term Aijlk and the quantitative and qualitative

    measures. Although, these models have been applied to the MOFL problem,

    they all have two inadequacies:

    1. All factors may not be represented on the same scale: for example,values for work flow may range from zero to a tremendous amount,

    ,ijlkRijlkCijlk RWCFA

    ,ijlkijijlk RDFA

    ,)(lkijijijlk DCRFA

    ,)( 21 lkijijijlk DRWFWA

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    33/143

    Chapter 3: Facility Layout Problem Formulation 21

    while closeness rating values may range from -1 to 4. Hence, the

    closeness ratings would be dominated by workflow and have little

    impact on the final layout Harmonosky and Tothero [8].

    2. Measurement units used for objectives are incomparable: the

    closeness rating represents an order preference indicating the

    necessity that given facilities be located close together. The total

    closeness rating score is only an ordinal value; on the other hand, the

    material flow handling is measured according to cost. Combining

    these two values with different measurement units in an algebraic

    operation is unsuitable Chen and Sha [21].

    For the reasons cited above, Harmonosky and Tothero [8] suggested an

    approach that normalizes all factors, before combining them. To normalize a

    factor, each relationship value is divided by the sum of all relationship values

    for that factor, as shown in equation (11).

    Where Sijmis the relationship value between departments i and j for factor m,

    and Tijm is the normalized relationship value between departments i and j for

    factor (objective) m.

    Next, all values are multiplied by weights representing the relative importance

    of each factor m. Then, the sum of all values for each departments pair is

    calculated.

    The resulting objective function is shown in equation (12).

    (11)

    N

    u

    N

    v uvm

    ijm

    ijm

    S

    ST

    s).(objectivefactorsofNo.Tandm,factorforwieghttheis

    (12). ),(),(

    m

    kjlilkijm

    N

    i

    N

    j

    M

    l

    M

    k

    T

    m m

    where

    XXDTQMin

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    34/143

    Chapter 3: Facility Layout Problem Formulation 22

    Harmonosky and Tothero [8] proposed a methodology for normalizing all

    factors into comparable units on the same scale. However, the scaling

    problem remains unresolved. Note first that values for work flow may range

    from zero to a very large positive value, while closeness rating values may

    range from a negative value to a positive value. After using equation (11),

    most normalized relationship values of larger scaling factor are lower than

    those of smaller scaling factor. As a result, the larger scaling factor would

    have very little effect on the final layout. Second, different scoring values for

    closeness ratings may cause some inadequacies as mentioned in Chen and

    Sha [21] and presented in the literature. However, Chen and Sha [21]

    proposed an approach that normalizes all factor before combining them. In

    order to achieve normalization, they subtract the mean of the layout cost

    distribution from each objective value and divide the result by the standard

    deviation of all feasible layout costs, as exhibited in equation (13).

    The variance expression Vm, has been proposed by Khare et al [77]. The

    meanMm of all feasible layout costs can also be computed Nugent et al. [41].

    Chen and Sha [21] considered only those objectives with distance-weightedattributes such as flow and closeness rating. For this reason, they proposed

    this approach for solving the MOFL problem, which is based on the

    minimization of distance-weighted objectives. This is achieved by minimization

    of total flow cost (TFC) and minimization of total numerical rating (TNR) that is

    presented by Khare et al. [76]. Hence, all distance-weighted objective

    functions can be characterized as a normal distribution. Using proposed

    equation (13), they reasonably normalize all objectives, and resolve both the

    different scale and measurement unit problems. The values obtained are then

    .objectiveforvaluenormalizetheis

    .objectiveforondistributicostlayouttheofvariancetheis

    .objectiveforondistributicostlayouttheofmean valuetheis

    )21.(objectiveforlocationatfacilityandlocationatfacilitylocatingofvalueobjectivetheis

    (13),2/1m

    mH

    mV

    mM

    ,...,t,mmkjliS

    where

    V

    MSH

    m

    m

    m

    ijlkm

    m

    m

    N

    i

    N

    j

    M

    k

    M

    l ijlkm

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    35/143

    Chapter 3: Facility Layout Problem Formulation 23

    multiplied by weights (Wm) representing the relative importance of each

    objective. In their proposed model, the resulting objective function is shown in

    equation (14):

    Owing to the proposed model of Chen and Sha [21], as mentioned, that has

    the superiority of the previous models; it is used as the evaluation criteria for

    the solution quality in the current work.

    (14). t

    m mmHWQMin

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    36/143

    Chapter 4

    A Review of Genetic Algorithms

    4.1 What is a Genetic Algorithm?

    enetic algorithms (GAs ) are a problem solving technique hinted at by

    living creatures evolution Whitley [78]. In GAs, chromosomes, linear

    encodings of a problems possible solution, are selected from a population,

    operations such as crossovers and mutations are applied; and they survive in

    higher probability if they are regarded as a better ones. That is, a GAs

    mechanism is similar to natures one in which superior individuals can

    produce more descendants in the future.

    In order to use GAs for solving a problem, important points are considered:

    Representation of the chromosomes;

    Design of crossover and mutation operators; and

    Fitness functions.

    G4

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    37/143

    C h a p t e r 4 : A R e v i e w o f G e n e t i c A l g o r i t h m s 25

    4.2 GA Parameters and Performance

    In GAs, there are many kinds of parameters, which influence the GAs

    behavior. The next section, some important GA parameters and their

    influences will be briefly reviewed.

    4.2.1 Crossover and Mutation: crossover usually takes two parents and

    can reproduce one, two or more children. In actual GAs, the allele of either

    parent is simply copied into the corresponding place of the childs

    chromosome. There are some variations e.g. one-point crossover, two-point

    crossover, and uniform crossover.

    One-point crossover first specifies a split point on a chromosome at

    random; then copies the alleles between the head and the splitting

    point of one parent and those between the splitting point and the tail of

    the other parent.

    Two-point crossover initially choose two splitting points at random; then

    duplicates the alleles between the splitting points of one parent, and

    the other alleles from the other parent.

    Uniform crossover randomly picks each allele from either of the two

    parents.

    If two parents 1-2-3-4-5 and 6-7-8-9-0 are selected, a child 1-2-8-9-0 may be

    created by one-point crossover, 1-7-8-4-5 may be produced by two-point

    crossover, and uniform crossover may generate 1-7-8-4-0.

    In contrast, mutation may happen to one allele or probabilistically to every

    allele. A chromosome 1-2-3-4-5 may be changed to 1-6-3-4-5.

    In GAs, crossover rate and mutation rate are applied. For instance, if

    crossover rate is .4 and mutation rate is .01, crossover will happen with 40%

    probability and mutation will occur on each allele with 1% probability.

    Therefore, if the chromosome consists ofL alleles, a chromosomes changing

    probability by the mutation is 1-(1-M)LLMwhereMis the mutation rate andM

    can be assumed to be less than 1/L Goldberg [79].

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    38/143

    C h a p t e r 4 : A R e v i e w o f G e n e t i c A l g o r i t h m s 26

    4.2.2 Population Size: the number of chromosomes is often called

    population size, and it may also influence the GA. As mentioned in Goldberg

    [79], a GA of large population size may have better solutions ultimately

    because of large number of chromosomes may include good schemata insome chromosome. On the other hand, GAs with smaller population can

    change rapidly; therefore, it may show better performance in the early stages

    rather than those with a larger population.

    4.2.3 Selection Methods: There are many kinds of selection methods; three

    methods are introduced here:

    Rank method Baker [80] first sorts all the chromosomes in order of fitnessvalues; then, the probability of selecting a particular chromosome is

    proportional to the inverse for the order rather than the fitness itself. If there

    are five chromosomes whose fitness values are 1, 6, 3, 8, 5; then the rank

    ordering is 5th, 2nd, 4th, 1st, 3rd and the probabilities of selection are 1/15,

    4/15, 2/15, 5/15, 3/15.

    In contrast, tournament selection Brindle [81] is as follows. First, a particular

    numberS is decided as the size of tournament. Second, S chromosomes are

    uniformly chosen from all the chromosomes (population size). Finally, the best

    one among the S chromosomes is selected as a parent. Of course, two

    parents are required in normal GAs; therefore, the above process is usually

    done 2N times, where N is the population size. In the tournament selection,

    the same chromosome may be chosen more than once Whitley [78].

    However, the tournament selection with large Scauses a strong pressure to

    choose very fit chromosomes. And, this leads premature convergence of

    chromosomes, which usually produce only poor solutions. To explain it, let us

    consider the probabilities of being chosen as a parent for the five

    chromosomes: the best one in the generation; the 75th percentile; the median;

    the 25th percentile; and the worse one. In tournament selection, each

    candidate for parents has to win against (S-1) competitors in a group to

    become a parent. Because the probabilities of meeting a weaker chromosome

    for the five chromosomes are 100%, 75%, 50%, 25% and 0%; the

    probabilities of becoming a parent by winning against (S-1) competitors for the

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    39/143

    C h a p t e r 4 : A R e v i e w o f G e n e t i c A l g o r i t h m s 27

    five are 1, (0.75)s-1, (0.5)s-1, (0.25)s-1 and 0. Thus, ifS= 2, they are 1, 0.75, 0.5,

    0.25 and 0; and ifS=5, they are 1, 0.32, 0.06, 0.004 and 0. Hence, we can see

    that very fit chromosomes will be frequently chosen as parents in largeS.

    Modified tournament selection method may be useful in some cases to

    cope with this defect as mentioned in Ross and Hallam [82]. In this method, a

    chromosome is first chosen as the first candidate at random; secondly, the

    chromosome is compared with at most (S-1) chromosomes randomly chosen.

    If a better chromosome than the first candidate is found from the (S-1)

    chromosomes, then the better one is selected as a parent immediately;

    however, if all the (S-1) chromosomes are worse than the first candidate, then

    the first one is selected as a parent. Therefore, other candidates than the first

    one can become the parent only by beating the first candidate. Hence, the

    strong pressure to choose very fit chromosome observed in tournament

    selection should become weaker in this modified tournament selection.

    4.2.4 ReproductionMethods there are some variations of reproduction

    methods. In the next section, two methods are explained:

    Genitor method Whitley [83], is considered as one of the ( + ) evolution

    strategy Baeck et al [84], in which offspring are produced from parents,

    and the best chromosomes of ( + ) are retained. Because the best

    chromosomes are always retained in this strategy, the population may

    converge gradually without drastic drifts. On the other hand, the generation-

    based GA is regarded as one of the (, ) evolution strategy, in which

    offspring are produced from parents and the best chromosomes of are

    retained. Because the best chromosomes may be lost in this strategy, the

    population may dramatically drift in search space for solutions. Therefore, all

    the chromosomes in Genitor may converge quicker than those in generation-

    based GA, Davis [85]; but Genitor may produce only poor solutions due to

    premature convergence Whitley [78]. For more information about genetic

    algorithms Schaffer et al. [86], Goldberg [79], and Holland [87].

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    40/143

    C h a p t e r 4 : A R e v i e w o f G e n e t i c A l g o r i t h m s 28

    4.3 On-line, Off-line and Best Individual Performance

    According to Baker [80], there are three types of criteria to evaluate the GA

    performance. They are:

    1. On-line performance, the average of all results that have appeared;

    2. The off-line performance, the average of best results of each

    generation; and

    3. The best individual performance, the best result that has appeared.

    While the off-line and the best individual performances only take account of

    the best chromosome in each generation, the on-line performance reflects the

    performance of chromosomes other than the best one as well. Therefore, the

    GAs showing good on-line performance may not produce remarkable

    chromosomes.

    Between the off-line and the best individual performances, the off-line

    one can take the convergence speed into account unlike the best individual

    performance. If two algorithms X and Y show the best individual performance

    as shown in Figure 3.1, the off-line performance of algorithm X is better than Y

    at time T1, though the best individual performance of both algorithms are

    same. But, at time T2, the off-line performance of algorithm X is still betterthan Y, although its best individual performance is worse. That is, off-line

    performance is generally influenced by the past records. However, because

    the quality of the best solution may be important in practical applications, the

    best individual performance may be more useful than the other performance

    measures. Hence, the best individual is used in the current thesis.

    Figure 3.1 Two algorithms performance.

    Time

    Fitness

    of

    the

    bestindividual

    T1 T2

    Algorithm Y

    Algorithm X

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    41/143

    C h a p t e r 4 : A R e v i e w o f G e n e t i c A l g o r i t h m s 29

    4.4 GA Application

    As Glover and Greenberg [88] mentioned, GAs may be effective

    approaches for NP-complete problems as ones of stochastic approaches. For

    example, Fang et al [89] and Ross et al [23] introduced GAs efficiency on job-

    shop scheduling problems and time tabling problems, respectively. Similarly,

    various layout problems and FLPs are solved by GAs. Here, some of them will

    be reviewed.

    4.4.1 Layout Problems with GAs: Regarding layout-related research,

    Kloske and Smith [24] tackled cable routing problems with a GA. In the GA, a

    chromosome consists of the index of each cables routing alternatives. For

    example, if there are three cables to be routed, a chromosome will have three

    alleles. And, if a chromosome is 2-3-2; the first cable will be routed by the

    second possible way for the cable, the second cable will be routed by the third

    possible way for the cable, and the third cable will be routed by the second

    possible way for the cable. Although Kloske and Smith [24] did not report any

    details how each cables alternative ways are produced, they state that the

    GA worked well.

    Whereas Kloske and Smith [24] were able to use traditional crossovers

    and mutations,while Smith [27] had to use a modified crossover to cope with

    his chromosome encoding in a bin-packing problem. In his research, a

    chromosome represents a list of packing order of objects. For instance, if a

    chromosome is 4-1-3-2; then it represents that object No. 4 will be first

    packed, object No. 1 will be second packed, and so on. Therefore, ordinary

    crossover may produce nonsense children. If 4-1-3-2 and 1-2-3-4 are one-point crossed over, and if splitting point is set between the second and third

    genes; then, it may produce twin children 4-1-3-4 and 1-2-3-2, which do not

    represent solutions of this problem. In order to tackle this problem, he used

    modified crossover, which keeps the genes before splitting point of the first

    parent and applies the order in the second parent for the rest of objects.

    Therefore, in the above example, 4-1-2-3 and 1-2-4-3 may be created instead.

    Furthermore, Falkenauer [28] took account of the redundancy of the

    representation of chromosomes. For example, in the encoding of Smith [27],

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    42/143

    C h a p t e r 4 : A R e v i e w o f G e n e t i c A l g o r i t h m s 30

    1-2-3-4 and 4-3-2-1 may be different solutions. Nevertheless, if objects 1 and

    2 fill a bin and if object 3 and 4 fill another bin, these two chromosomes

    virtually represent the same solution. Because higher redundancy makes the

    GAs search space larger and GAs power weaker, he suggested another

    encoding method to reduce the redundancy. Although it might be highly

    dependent on the problems domain, we may be able to see the fact that

    chromosomes representation and design of crossovers and mutations will

    much influence GA performance.

    4.4.2 FLPs with GAs: As regards FLPs, various kinds of chromosome

    representations and crossovers and mutations have been also suggested.

    Cohoon and Paris [90] introduced an original crossover method for the cell

    assignment representation. In a 3x3 FLP, a chromosome 1-2-5-6-7-9-3-8 may

    represent a layout shown in Figure 3.2. Since the conventional crossover

    tends to favor shorter schemata more Whitley [78], the relation of facilities No.

    5 and No. 6 may be kept in higher probability than that of No. 2 and No. 6, in

    this example. Cohoon and Paris [90] introduced another special crossover

    which can take into account such two dimensional adjacencies.

    1 2 3

    5 6 7

    4 8 9

    Figure 3.2 Chromosome representation for an identical FLP.

    Smith and Tate [91] used flexible bay structure representation to tackle non-

    identical FLPs. In the representation, a physical layout is represented by two

    chromosomes. The first one specifies the order of putting facilities into cells,

    and the second chromosome specifies how many cells are included in each

    bay (row). Smith and Tate [91] used the crossover as follows: the childs first

    chromosome is produced from the first chromosomes of parents by the same

    method of Smith [27]; and the childs second chromosome is copied from

    either parents second one. As for the mutation, one of the following three

    types is applied. If the first type (MU1) is applied, a bay chosen at random is

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    43/143

    C h a p t e r 4 : A R e v i e w o f G e n e t i c A l g o r i t h m s 31

    divided into two bays. If the second type (MU2) is applied, two sequential bays

    chosen at random are merged into one bay. If the third type (MU3) is applied,

    a part of genes will be reversed. That is, MU1 and MU2 affect the second

    chromosome, whereas MU3 affects the first chromosome. Smith and Tate [91]

    set the probability ratio of MU1, MU2, and MU3 occurring to be 1:1:2.

    While the flexible bay structure requires each cell to lie in rows (bays),

    the slicing tree structure (STS) can generate more various physical layouts.

    However, if Polish expression, which corresponds to a tree structure ant to a

    layout, is directly used as a chromosomes representation, ordinary

    crossovers and mutations can not be applied, because a combination

    operators and operands in random order may not be a valid Polish expression

    (solution).

    In order to use Polish expression as chromosomes representation,

    Cohoon et al [25] suggested several types of special crossover and mutation

    methods. For example, one of crossover methods creates a child so that it

    can inherit the trees structure from one parent and that it can inherit the

    operators in the Polish expression from the other parent. As for mutations,

    Cohoon et al [25] used Wong and Lin [92]s methods, which are used, for

    solutions move in simulated annealing. They are, swapping adjacent

    operands; switching a sequence of adjacent operators; and swapping an

    operator and a neighborhood operand.

    On the other hand, in order to use conventional crossovers and mutations,

    Tam [26] suggested a method where the tree structure is fixed and a

    chromosome includes only operators of the Polish expression. In other words,

    Tam [26] limited the search space, while Cohoon et al [25] did not.

    In conclusion, there have been many methods for representation,

    crossovers, and mutations; and this might suggest that better methods may

    appear in the future. The proposed GA and its operators will be explained in

    details in the next chapter.

    4.5 The Need for Good Solver Techniques

    From the previous chapter, it is seen that the QAP is capable of

    representing the inter-cell layout problem, but the complexity of the situation

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    44/143

    C h a p t e r 4 : A R e v i e w o f G e n e t i c A l g o r i t h m s 32

    appear from the fact that the QAP formulation belongs to the class of NP-

    complete problems as it is cited in Garey and Johnson [12], and no definite

    method can arrive at an optimal solution in a reasonable time when 15 or

    more facilities are considered. Consequently, many heuristic algorithms have

    been developed for achieving a trade-off between computation time and the

    efficiency of the final solution Kusiak and Heragu [13].

    In search for effective solution procedures, the Genetic algorithm, an offshoot

    of the artificial intelligence (AI) submerged as an effectual tool capable of

    treating the NP-hard FLP problem.

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    45/143

    Chapter 5

    Proposed Genetic Algorithm

    5.1 Why Genetic Algorithm?

    GAs differ from traditional optimization and search procedure in fourways Davis [85]:

    They work with a coding of the parameter, not the parameters

    themselves;

    Start search from a set of population points, not single point;

    Use payoff information, not derivatives or other auxiliary knowledge; and

    Use probabilistic transition rules, not deterministic rules.

    5.2 The Proposed Genetic Algorithm

    In the following subsections, the proposed genetic algorithm variables

    will be discussed in details. Such chromosome representation, initial

    population, selection method, crossover and mutation operator, and

    reproduction systemetc.

    5

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    46/143

    C h a p t e r 5 : P r o p o s e d G e n e t i c A l g o r i t h m 34

    5.2.1 Chromosome Representation:

    Chromosome representation maps feasible solutions of the problem.

    The effectiveness of the crossover operator depends greatly on the

    representation scheme used. The representation should be such that the

    crossover operator preserves high performing partial arrangements

    (schemata) of strings, and minor changes in the chromosome translate into

    minor changes in the corresponding solution.

    From the three FLP representation methods, mentioned in the previous

    sections, the cell assignment representation (permutation method) is used to

    represent the facilities layout, such that the allele ( j ) at position ( i ) in the

    genome indicates that facility ( j ) is assigned to location ( i ). For example,Individual 3,4,7,1,8,2,9,6,5 represents a solution for the shown layout in

    Figure 5.1.

    Figure 5.1 Phenotype representation layout.

    5.2.1.1 Representation of special layout configuration

    First, if the layout area has some restricted location, at this situation,

    each restricted location in the layout is encoded to 1 in the genotype. For

    instance, Figure 5.2 is represented by this schema:(#,#,#,#,#,#,#,-1,#,#,-1,-1,-

    1,#,#).

    Figure 5.2 Available layout space with restricted area.

    Where #: means any alleles can be assigned to this position except what has

    assigned before.

    Note that all chromosomes have this schema.

    3 4 7

    1 8 2

    9 6 5

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    47/143

    C h a p t e r 5 : P r o p o s e d G e n e t i c A l g o r i t h m 35

    Second, how can represent the chromosome if there are any pre-specified or

    occupied facilities. Figure 5.3 is represented by this schema:

    (#,#,7,#,#,#,#,#,3,#,-1,-1,-1,#,#).

    7

    3

    Figure 5.3 Available layout space of restricted area

    and pre-specified facilities.

    Note that all chromosomes have this schema.

    5.2.2 Ini t ia l Population:

    The initial population of chromosome is generated randomly. For not

    generating illegal chromosome, each process of generating a random gene

    (facility) checks the previous genes (facilities), which are once picked in the

    previous generations and choose a gene that has not been chosen before in

    the chromosome. This procedure ensures that each gene is only once chosen

    in a chromosome. This initialization keeps proceeding until the number ofchromosome reaches the pre-set population size (PopSize). In the current

    study, PopSize considered is 200.

    5.2.3 Select ion Method

    The selection criterion is used to select the two parents to apply the

    crossover operator. The appropriateness of selection method for a GA

    depends upon the other GA operators chosen. In the literature, a typical

    selection method gives a higher priority to fitter individuals since this leads to

    a faster convergence of the GA. Nevertheless, the overall results are obtained

    when the selection is not biased towards fitter individuals. So, the tournament

    selection method is applied in order to control convergence speed by the

    tournament size, Ts. In the current study, Ts considered is 4.

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    48/143

    C h a p t e r 5 : P r o p o s e d G e n e t i c A l g o r i t h m 36

    5.2.4 Crossover Operator

    The crossover operation corresponds to the concept of mating. It is

    hoped that the crossover (mating) of good parents may produce good

    offspring. Thus, the crossover operation is a simple yet powerful way of

    exchanging information and creating new solutions. Partially matched

    crossover (PMX) [93] operator is employed with probability Pc here to

    generate two offspring. The PMX operator starts as follows, at first two cut

    points are chosen at random for the parents. Then the genes of the father

    string bounded by the cut points will be copied to the same positions of the

    first offspring, and the remaining genes of the offspring will be filled up by the

    mother string in the same order. In such case, the offspring may not be legalbecause of possibilities of repeated genes. In this case, the repeated genes

    will be replaced by genes corresponding to the mapping of the father and

    mother spring bounded by the cut points. Suppose the father is (2, 4, 5, 3, 8,

    9, 6, 1, 7), the mother is (3, 9, 8, 6, 5, 4, 2, 7, 1) and the cut points lie after the

    third gene and before the seventh gene. Thus, the father genes inside the cut

    points are 3, 8, 9, the mother genes outside the cut points are 3, 9, 8, 2, 7, 1.

    The offspring is (3, 9, 8, 3, 8, 9, 2, 7, 1). This would be illegal because 3, 8, 9,

    are repeated genes. Note that 3, 8, 9 of the father genes map to 6, 5, 4, thus

    we replace 3 by 6, 8 by 5, and 9 by 4, to get the new offspring as (6, 4, 5, 3, 8,

    9, 2, 7, 1). The second offspring is generated by the same manner but it uses

    the mother genes inside the cut points, and the remaining genes are filled up

    by father string in the same order. Figure 5.4 shows the PMX method to

    generate feasible solutions. In the current study,Pc considered is 0.7.

    Cut points

    Father 2 4 5 3 8 9 6 1 7

    Mother 3 9 8 6 5 4 2 7 1

    First offspring 2 4 5 6 5 4 6 1 7

    Second offspring 3 9 8 3 8 9 2 7 1

    Legal 1st

    offspring 2 9 8 6 5 4 3 1 7

    Legal 2nd

    offspring 6 4 5 3 8 9 2 7 1

    Figure 5.4 PMX crossover method.

  • 7/28/2019 10381067 . Solution Methodology Based Genetic Algorithm for Multi-Objective Facility Layout Problem

    49/143

    C h a p t e r 5 : P r o p o s e d G e n e t i c A l g o r i t h m 37

    5 .2 .5 Mutation Operator

    If the entire population has only one type of string (or strings are verysimilar), then the crossover of two strings does not produce any new strings.

    To escape from this scenario, the mutation operator may be used. In our GA,

    The mutation operation may happen with probability (Pm, mutation rate), to

    swap every allele (gene) and selected allele at random (order change

    mutation). For instance, chromosome 1, 5, 6,3, 4, 2 may be change to 1, 3, 6,

    5, 4, 2. In the current study,Pm considered is 0.1.

    Note that pre-specified or occupied facilities and/or restricted locations are not

    affected by the crossover and/or mutation operators.

    5.2.6 Reproduction System

    The generation-based system is used. That is, offsprings from

    parents are produced and the best chromosome ofare retained. Owing to

    making new population only by new offspring can cause lost of the best

    chromosome from the last population. This is true, so called Elitism is often

    used. This means, that at least one best solution is copied without changes t