Você está na página 1de 24

Pareto Optimal Based Evolutionary

Approach for Solving Multi-Objective


Facility ayout Problem
!a"i Shah #a$a" %ipon& !yrre 'lette& Omid
Mirmotahari& Mats ()vin& *im +)rresen
,epartment of -nformatics& .niversity of
Oslo& #or$ay
Facility ayout Problem
/FP0

FPs deal $ith arranging a given number of facilities


/departments0 on the factory 1oor of a manufacturing system to
meet one or more objectives2

+hese objectives may re1ect material handling costs or


preferences regarding adjacencies among departments2

-n essence& FP can be considered as a searching or optimi"ation


problem& $here the goal is to 3nd the best possible layout2

+raditionally FP has been presented as a 4uadratic Assignment


problem /4AP0 to 3nd the best assignment of n facilities to m
locations& $here the number of facilities and locations must be
e5ual2
Multi-Objective FP

Over the years& various evolutionary approaches have been proposed in


e6orts to solve the facility layout problem /FP02

.nfortunately& most of these researches are generally concerned $ith a


single objective& either qualitative or quantitative feature of the layout2

By contrast& practical layout problems are multi-objective by nature and


they re5uire the decision ma7ers to consider a number of criteria
involving both quantitative and qualitative objectives before arriving at
any conclusion2

A solution that is optimal $ith respect to a given criterion might be a


poor candidate for some other criteria2

(ence& the trade-o6s involved in considering several di6erent criteria


provide useful insights for the decision ma7ers2

Surprisingly& the research in this important 3eld has been scarce $hen
compared to the research in single criterion2
4uantitative Objectives

4uantitative approaches
involve primarily the
minimi"ation of material
handling costs bet$een
various departments2

More speci3cally& it tries to


minimi"e the sum of
distance among all facilities
multiplied by the
corresponding 1o$s2

+he material 1o$ is


e8pressed as the from-to
chart2
Figure9 From-to :hart used in 5uantitative
objective
4ualitative Objectives

Evaluating facility layouts based on the


distance-based objective is often not enough2

+he 5ualitative aspects of layouts may also


need to be considered2

4ualitative goals include placing departments


that utili"e common materials& personnel& or
utilities adjacent to one another& $hile
separating departments for reasons of safety&
level of $or7 1o$& technological reasons& user;s
preference& noise& or cleanliness2

4ualitative approaches used the closeness


rating scores to indicate the desired relative
<closeness= re5uirement for t$o departments
to be ne8t to each other2

(ere the goal is to ma8imi"e the total closeness


ratings among all departments2

+he closeness ratings are> A /absolutely


necessary0& E /essentially important0& -
/important0& O /ordinary0& . /un-important0 and
? /undesirable0& to indicate the respective
degrees of necessity that any t$o given
departments be located close together2
Figure9 :loseness rating
chart
Figure9 4ualitative
objective
@eighted Sum Method

Multiple objectives are combined into a single scalar objective using $eighted coeAcients2
Disadvantages:

%elative $eights of the objectives are not e8actly 7no$n in advance and cannot be pre-
determined by the users&

objective function that has the largest variance value may dominate the multi-objective
evaluation2

as a result& inferior non-dominated solutions $ith poor diversity $ill be produced2

.ser al$ays has to specify the $eight values for functions and sometimes this $ill not
have any relationship $ith the importance of the objectives2

A single solution is obtained at one time2

-f $e are interested in obtaining a set of feasible solutions& it has to be run several times2

this also& is not a $arranty that the solutions obtained in di6erent runs are di6erent2

More importantly& since the selection of objective $eights is critical in designing layout
having multiple objectives& the objective $eights therefore play an important role in the
design process2

-n practice& it is selected randomly by the layout designer based on hisBher past


e8perience that restricts the designing process completely designer dependent and thus
the layout varies from designer to designer2

+o overcome such diAculties& Pareto-based evolutionary optimization has become an


alternative to classical $eighted sum method2
Pareto ,ominance

-n the totally absence of information for preferences of the


objectives& Pareto dominance is regarded as an appropriate
approach to compare the strength bet$een t$o solutions in
multi-objective optimi"ation2

-n other $ords& for problems having more than one objective


function& any t$o solutions x
(1)
and x
(2)
can have one of the t$o
possibilities- one dominates the other or none dominates the
other2 A solution x
(1)
is said to dominate the other solution x
(2)
& if
both the follo$ing conditions are true9
C0 +he solution x
(1)
is no $orse than x
(2)
in all objectives2
D0 +he solution x
(1)
is strictly better than x
(2)
in at least one
objective2
Pareto ,ominance

+he set of all such solutions


$hich are non dominated
constitute the Pareto front2
Figure : Pareto front
'oals of Multiobjective Optimi"ation
C0 +o 3nd a set of solutions as close as possible to the Pareto
optimal front2
D0 +o 3nd a set of solutions as diverse as possible2
-mportance of Multi-
Objective FP

(istorically& FPs have been solved only for one goal& either 5uantitative or
5ualitative aspect of the layout2

-n real-$orld FPs& it is often necessary to optimi"e both 5uantitative and


5ualitative criteria simultaneously2

-n general& minimi"ation of the total material handling costs is often used as


the optimi"ation criterion2 (o$ever& closeness rating& ha"ardous movement
or safety& and the li7e are also the important criteria in FP2

Many researchers have 5uestioned the appropriateness of selecting a single-


criterion objective to solve FP because 5ualitative and 5uantitative
approaches each have advantages and disadvantages

Accordingly& it is desirable to generate many near-optimal layouts


considering multiple objectives according to the re5uirements of the
production order or customer demand2

the production manager can selectively choose the most demanding one
among all of the generated layouts for speci3c order or customer demand2

Also& it is desirable to obtain as many di6erent Pareto-Optimal layout


solutions as possible& $hich should be non-dominated& converged to& and
diverse along the Pareto-optimal front $ith respect to these multiple criteria2
Proposed Pareto Optimal based
Approach
Figure. Flowchart of multi-objective FLP approach
:hromosome
%epresentation

A form of direct representation for strings is used2

+he solution is represented as a string of integers of length n& $here n is


the number of facilities2

+he integers denote the facilities and their positions in the string denote
the positions of the facilities in the layout2
E C F G H I J K D
Figure: Layout
Figure: Chromosome representation of layout
Figure: Chromosome representation
Objective Functions
/5uantitative0

+he 3rst 3tness function& total material handling cost& is based on


5uantitative model2

+his function is subject to minimi"ation& and measured as


Objective Functions
/5ualitative0

+he second 3tness function& the closeness rating score& is based on


5ualitative model2

+his function is subject to ma8imi"ation& and measured as


:rossover

@e follo$ the concept described by Suresh et2 al LCM for crossover2

-t maintains partial structure of the parents to a large e8tent than


by the e8isting crossover operations such as PM?& O?& :? LCM2
P1 = 1 2 3 | 4 5 6 7
8
P2 = 4 7 1 | 5 2 8 3
6
After crossover
A1 = 1 2 3 | 6 3 8 2
5
A2 = 8 7 6 5 4 | 4 7
1
After repair
C1 = 1 7 4 6 3 8 2 5
C2= 8 2 6 5 3 4 7 1
[1] Suresh, G., Vinod, V.V., Sahu, S.: A Geneti Algorithm for Faility Layout. !nt. ". #rod. $es. %%&1'(, %)11*%)'% &1++,(
Mutation

+$o genes are pic7ed up randomly and then they e8change their
positions2

As a result& the resultant chromosomes are legal and no repair is


re5uired2
E8perimental %esults and
,iscussion

.ntil no$& almost all FP algorithms try to optimi"e single


criteria only /mainly minimi"ing the material handling cost02

+hough fe$ multi-objective FP approaches can be found in


the literature& all of these e8isting approaches employed
$eighted sum approach

+herefore& to evaluate our proposed algorithm& 3rst $e


compared the material handling costs obtained by our
approach $ith the e8isting single objective approaches to
justify its capability to optimi"e this cost2

+hen $e demonstrated its performance as a multi-objective


evolutionary FP algorithm by optimi"ing material handling
cost and closeness rating score2

#ote that& for both cases $e have used the same results
achieved by our approach2
Benchmar7 Problems

+o evaluate the proposed multi-objective FP approach& $e


run the algorithm on various benchmar7 data collected from
e8iting literature and $eb2

+he problems composed of I& J& E& CD& CF& DN& GN& HD& KD and
CNN facilities2

Oery fe$ benchmar7 problems are available for multi-


objective FP& particularly& in the case of closeness rating
score2

+hus& the authors have themselves created their o$n data


sets for closeness rating score2
E8perimental Setup
#opulation : ,- hromosomes &up to 1, failities pro.lems(
: 1-- hromosomes &a.o/e 1, failities pro.lems(
Generations : )- &up to 1, failities pro.lems(
: 0- &a.o/e 1, failities pro.lems(
Crosso/er rate : -.+
1utation rate : -.%

Each benchmar7 problem is tested for thirty times $ith


di6erent seeds2

+hen each of the 3nal generation is combined and a non


dominated sorting LCM is performed to constitute the 3nal
non dominated solutions2
[1] 2e., 3., #ratap, A., Agar4al, S., 1eyari/an, 5.: A Fast and 6litist 1ultio.7eti/e Geneti Algorithm: 8SGA9!!. !666
5ransation on 6/olutionary Computation :&'(, 10'*1+; &'--'(
Single Objective :onte8t
Table 1. Comparison 4ith e<isting algorithms
Single Objective :onte8t
Table 2. :omparison of the material handling cost obtained by various
evolutionary methods
Multiple-Objective
:onte8t
Table 3. $esults of 5est #ro.lems
Multiple-Objective
:onte8t

+o illustrate the convergence and


diversity of the solutions& non-
dominated solutions of the 3nal
generation produced by the
proposed algorithm for the test
problems naug30 and sing! are
presented in Fig2

From these 3gures& it can be


observed that the 3nal solutions
are $ell spread and converged2

And for this reason it is capable of


3nding e8treme solutions2

-n the 3gures& the occurrences of


the same non-dominated
solutions are plotted only once2
&a( naug%-
&.( singh:
Fig. 1. Final #areto9optimal front
Multiple-Objective
:onte8t

+hese 3gures demonstrate the


convergence behavior of the
proposed methods over
generations for ds"2

+hese 3gures also justify that our


proposed approach clearly
optimi"es both of the objectives
$ith generations2

From the 3gures& it can be found


that from 3rst generations to last
generations& the proposed
method is able to optimi"e both
of the material handling cost
/minimi"e0 and closeness rating
score /ma8imi"e02
&a( 1aterial handling ost
&.( loseness rating sale
Fig. 54o o.7eti/es o/er generations of ds8 pro.lem
Summary

+he proposed approach is capable of producing near optimal


and non-dominated /Pareto optimal0 solutions& $hich are also
the best-7no$n results in many cases2

+he simulation results clearly sho$ that our proposed approach


is able to 3nd a set of diverse Pareto optimal solutions& $hich
ful3lls the t$o main goals of the multi-objective FP algorithm2

Você também pode gostar