Академический Документы
Профессиональный Документы
Культура Документы
Vol. 17/No. I
1998
and the tabu search algorithm. Kouvelis and Chiang6 gantries. These devices may be different physically
used SA for solving a machine layout problem with but share one common attribute. They usually have
a straight material handling track so that the total to follow some fixed paths for material transporta-
backtracking flow distance can be minimized. tion; ~5 for example, AGVs need to follow the guide
Kouvelis, Chiang, and Fitzsimmons 7 used SA for paths laid on the floor. Other important issues of the
machine layout problems not confined to single row modeling framework are addressed in the following
problems under zoning constraints with the objec- subsections.
tive of minimizing the sum of assignment cost and
material handling cost. Zoning constraints restrict 3.1 Evaluation Criteria
the arrangement of machines. Kusiak and Parks A common criterion for evaluating machine lay-
developed two methods for linear single-row outs is the total flow distance. The goal is to mini-
machine layout and linear double-row machine lay- mize the total flow distance. However, minimizing
out problems with an assumption that the flow path total flow distance may not be the most important
is a bidirectional straight line, and an objective func- goal for some situations. The material handling
tion of minimizing the total travel distance. As device employed and the characteristics of the flow
pointed out earlier, almost all of the methods path determine what criterion is important for a lay-
reviewed above are intended for the layout problem out design. For example, for a unidirectional straight-
with a straight-line flow path. In addition, none of line conveyor, one may want to minimize the number
them consider the effects of flow path characteristics of backtracking movements or total backtracking
on the layout design, such as the effects of flow path flow distance because backtracking movements can-
characteristics on the selection of evaluation criteria. not be achieved by the unidirectional conveyor alone,
Another related problem is FLA, which determines as explained in Section 1. On the other hand, if the
the relative positions of machines based on the oper- conveyor is bidirectional, minimizing total flow dis-
ation sequences of products. The objective is to tance may be more important and appropriate
design a flow line in which the number of backtrack- because backtracking movements no longer exist in a
ing movements is minirraz"ed. Early FLA methods bidirectional flow path.
(such as those by Noy,9 Singleton,1 Moore, u and Four evaluation criteria are provided in the pro-
Hollier12) are for single-machine-type problems posed layout procedure. The first is the minimiza-
where only one machine of any type is allowed in a tion of the number of backtracking movements. This
flow line. This type of problem is not realistic because criterion is most suitable for problems with unidi-
duplicate machines are very common in actual manu- rectional nonloop flow paths. The second criterion is
facttaing environments. A more realistic problem that the minimization of total backtracking flow dis-
allows duplication of machines was studied by tance. Similarly, it is most suitable for unidirection-
Carrie? 3 Other studies on this problem can also be al nonloop flow paths. The third criterion is the max-
found in Aneke and Carrie ~ and H o , Lee, and imization of the number of in-sequence movements.
Moodie? 4 The study by Ho, Lee, and Moodie also In-sequence movements are especially favorable in
suggested a network flow analysis. Unlike traditional the mass-production assembly or transfer-line sys-
FLA, network flow analysis does not restrict the flow tem. Finally, the fourth criterion is the minimization
path configuration to a linear configuration. Flow of total flow distance.
branching and skipping are allowed if they can be jus- All four criteria can be applied to unidirectional
tified. Although FLA does not provide exact positions nonloop flow paths. For bidirectional flow paths and
of machines, the solution it f'mds often serves as a unidirectional loop flow paths, only minimization of
good guideline for a detailed layout of machines. the total flow distance and maximization of the num-
ber of in-sequence movements can be applied.
3. Modeling Framework Minimization of the number of backtracking move-
It is assumed in this paper that automated materi- merits and minimization of the total backtracking
al handling devices are employed in the layout prob- flow distance cannot be applied to bidirectional prob-
lem. Popular automated material handling devices lems because backtracking movements do not exist in
include AGVs, robots, conveyors, and overhead bidirectional problems. As for the unidirectional loop
Journal of Manufacturing Systems
Vol. 17/No. 1
1998
flow path, backtracking is not necessary since one can given. In the example, the travel distance from M6 to
move from one machine to any other machines on the M2 is equal to 19, while the travel distance fromM2 to
unidirectional loop flow path without violating the M6 is 29. However, the Euclidean distances for both
feasible flow direction. As a result, backtracking will M6 to M2 and M2 to M6 are 9.43, and the rectilinear
not be allowed in the unidirectional loop flow path. distances are 13. These distances are different from the
The applicability of these criteria in different problem actual travel distance. In this paper, the travel distance
scenarios is staumarized in Table 1. In addition to of a movement is calculated as the length of flow path
considering each of the above four evaluation criteria segment traversed by the movement, which is the actu-
independently, one or more of these criteria may be al travel distance.
considered simultaneously. This can be done through
the use of a multiple-criterion objective function, 3.3 Minimum Safe Distance Between Machines
which will be addressed in Section 4.2.1. It is necessary to save some distance between
neighboring machines for the safe operation of
3.2 Travel Distance Calculation machines, the storage of work-in-process (WIP), the
Another important issue is how to obtain the correct movement of personnel, and many other purposes.
travel distance. Two popular distance calculation One assumption here is that each machine has its own
methods are the Euclidean distance method and the minimum required safe distance to be away from
rectilinear distance method. For both, the distance other machines. With this assumption, the length of
from a point i to another pointj is equal to the distance the flow path segment between two adjacent
fromj to i. However, in reality, the travel distance from machines is made to be equal to the maximum of the
i toj is not necessarily equal to the travel distance from minimum required safe distances of these two
j to i. In addition, Euclidean and rectilinear distances machines. If one also assumes that the I/O point of a
are not necessarily equal to the actual travel distance. machine is at the midpoint of the flow-path-neighbor-
For example, as shown in Figure 3, six machines are ing side of the machine, then the travel distance from
arranged along a unidirectional loop conveyor and the machine A to machine B can be obtained as follows:
feasible flow direction is clockwise. The coordinates
of the input/output (I/O) points of machines are also
DISA,B = Z0.5o(WD, + Y D j ) + m a x { S i , S j } (1)
(i,j)GNMA.B
(o,8) (3,8) (~) 03,8) - (16,8)
I - where:
A ........................ ~- !
DISA,B travel distance from machine A to machine B
i
!
/
WDt width of machine i
v
NMA,B a set containing every pair of adjacent
(o,o) (3,0) (8,0) (13,0) (16,0) machines along the route from machine A to
machine B
Figure 3
Example for Calculation o f Travel Distance St minimum required safe distance of machine i
Table l
Applicability of Criteria
Problem Scenario
Unidirectional, Bidirectional, Unidirectional, Bidirectional,
Evaluation Criterion Non-Loop Non-Loop Loop Loop
4
Journal of Manufacturing Systems
Vol. 17/No. 1
1998
One may argue that this strategy does not guaran- Second, SA has a reputation as a very effective heuris-
tee nonoverlapping of machines if the flow path is a tic optirniTation technique. Finally, Johnson et al: 6
loop. This is true only when machines are arranged have shown that starting at good initial solutions rather
along the inner side of the loop. Overlapping will not than randomly selected solutions in SA results in bet-
occur if machines are arranged along the outer side ter chances of producing good-quality solutions.
of the loop. This can be realized from Figure 4. In the remainder of this section, six F L A methods
Machines are arranged along a straight flow path as will be introduced. Each of these will be illustrated
in Figure 4a. The distance between two adjacent with an example problem. Then, issues pertaining to
machines is equal to the maximum of their minimum the application of SA to the problems investigated
required safe distances. It is also assumed that the here will also be addressed. These issues will include
safe distance has been saved for the first machine and the construction of a normalized weighted multiple-
the last machine on the flow path. If one folds the criterion objective function to be used in SA, the
line upward, connects both ends, and makes it into a proper setting of initial and freezing temperatures,
loop, then the machines may collide with each other and four different temperature reduction strategies.
(Figure 4b). This collision can be prevented if the
line is folded downward (Figure 4c). The same argu- 4.1 Phase One---Flow Line Analysis
ment can also be applied to a U-shaped flow path. As Four of Hollier's12 F L A methods (Methods I, 2, 3,
for the serpentine flow path, it is recommended that and 4) are modified and two new F L A methods
machines be arranged as in Figure ld; however, one (Methods 5 and 6) are proposed. Hollier's original
needs to reserve enough space to prevent the methods cannot be directly applied to the problems
machines within each turn from overlapping the flow investigatedhere because they are designed for single-
path segment at the opposite side. machine-type problems in which duplication of
machines is not allowed. The inapplicability of
4, Proposed Layout Procedure HoIlier's methods to the problems investigated here
The proposed layout procedure has two phases. can be fixed by giving every machine a unique identi-
Figure 5 shows the flow chart of the proposed two- fication number and Izeating each machine as if it
phase layout procedure. As shown in the figure, before were the only one of its kind. By doing so, one can
the two-phase procedure can be started, the flow path treat any problems as single-machine-type problems
characteristics and the evaluation criteria have to be and apply Hollier'smethods to them. Another modifi-
detem-dned. At the f'n~t phase of the two-phase layout cation of HoIlier'smethods isthat allflow interactions
procedure, an FLA method is selected and applied to
obtain a machine layout. SA then uses the layout as an
initial solution and improves it in the second phase.
The combination of FLA and SA is due to the follow-
ing reasons. First, it is well known that FLA methods
a ovakmt~ crtt~a
are capable of producing good flow-line designs.
I PMse0M-use~ it to L~
S ~ c t an FLA rneh'~ ~
Ansbu~
obtsin a layout of machines on
the flow ~ .
~ T ~ - ~ t~m ~ m n ,
~ e m n ~e ~auon un~ ~ ~ me ~ n g tem-
~ u ~ , and me a~pm~tn ~ stmgy
{I)) (c) ~ e g~e l~tout ~und at Rme Ore u k Ini~l ~uUon anO a ~ y SA
w ~ ~e abovesetting to obtaln a better solut~t.
F/gRre 4
Examples for Preventing Machine CoHisJon
(a) machine layout along a straight-line flow path, (b) machine Figure 5
collision occurs, and (c) machine collision prevented Flowchart of Two-Phase Layout Procedure
Journal of Manufacturing Systems
Vol. 17/No. I
1998
Table 2
will now be considered regardless of how insignificant From-To Matrix of the Seven-Machine Example Problem and the
they may seem. The original methods of Hollier allow Summed "From" and "To" Values of Every Machine
designers to ignore some flow interactions if they
seem insignificant to the designers. Hollier's intention To
From l 2 3 4 5 6 7 Row Sum
was to simplify a problem by reducing its size; how-
ever, because determining the significance of flow 1 0 2 3 4 5 6 1 21
interactions is very subjective, the original Hollier's 2 5 0 3 2 I 4 2 17
methods run the risk of sacrificing the quality of solu- 3 3 4 0 1 2 5 1 16
4 2 1 5 0 3 I 2 14
tions. The modified HoUier's methods and the pro- 5 1 2 3 2 0 1 4 13
posed methods are explained in the following sections. 6 3 2 3 3 1 0 1 13
7 2 3 1 5 4 2 0 17
4.1.1 Method 1
Col. Sum 16 14 18 17 16 19 11
This method tries to minimize the number of
backtracking movements. Before one can use this
method, a From-To matrix needs to be prepared. The
Inlletmmll~I~su# I
total "from" and "to" values of each machine also ~Wlru~d Row Line
need to be calculated. The flow line is then formed 7
the same end of the flow line, then the priority is UpdatedSummed"From" and Summed"To"
given to the one with the largest ratio of summed
"from" and summed "to", or the one with the
largest ratio of summed "to" and summed
ConstructedFlow Line
"from".
O J
of this example problem are illustrated in Figure 7. Updated"SummedTo/SummedFrom" and "Summed From/SummedTo"
ID2 3 5
mmea From/summedTo 0.667 1.0
4.1.3 Method 3
Intlt'lMdklte Relmlt V
This method tries to minimize the number of back- ConstructedRow Une
tracking movements. To implement Method 3, a sec- 7 1 4 5 6
ondary matrix must be prepared first. The new cell Updated"SummedTo/SummedFrom" and "Summed From/SummedTo"
MachineU~2 3
value, So, in the secondary matrix is equal to X u - Xz/, Sunllnl~ Tq/YJomml~F'rgm 1.333 0.75
where X.~/and Xj~ are the cell values in the original Sumrned R'mVStg~medTo 0.75 1.333
Table3
Initial " S u m m e d To/Summed From" and " S u m m e d F r o m / S u m m e d To" Ratios of Each Machine of the
Seven-Machine Example Problem Using Method 2
Machine ID 1 2 3 4 5 6 7
Summed To/Summed From 0.762 0.824 1.125 1.214 1.231 1.462 0.647
Summed From/Summed To 1.313 1.214 0.889' 0.824 0.813 0.684 1.545
7
Journal of Manufacturing Systems
Vol. 17/No. 1
1998
Table4
Secondary Matrix of the Seven-Machine
I ~ 1 1 ~ IQ 1 2 3 4 5 6 7
Example Problem for Method 3 of PmilNM/mollR ~ M ~ 0.444 0.5 1.667 1.6 2.5 4.0 0
/~=l=lulelkcnofNlmlwSumeflZml~ 2.25 2.0 0.6 0.625 0.4 0.25 -
MC 1 2 3 4 5 6 7 IIMIIBUlfB Rllult I
Commctel Row Line
7
1 0 -3 0 2 4 3 -1
2 3 0 -1 1 -I 2 -1 L~--"~--~ = of "Nelmve/Po=~" and "Pee~WNe0=M"
3 0 1 0 -4 -1 2 0 II/ll~ ~ 1 2 3 4 5 6
,Samof P ~ b W ~ Smof~,~ 0 . ~ o.~m 1.set ~.0 2.5 3.s
4 -2 -1 4 0 1 -2 -3 ~ S m ~ o f P e . ~ 3,o S.0 0.S ~.0 0.4 O.286
5 -4 1 1 -1 0 0 0
lulmmml~ Itlmlt II
6 -3 -2 -2 2 0 0 -1 Cmmn~l Row Line
7 1 1 0 3 0 1 0 7 6
Upe== ~ of "Ne~va/P=~" =W "~WNega~ve"
Table5
Initial Absolute Sum of"Negative" and Sum of"Positive" of the MIICI1NID 1 2 3 4 5
SumV,~m/~Swaof~ 0,~ 0.~ 5.0 0.e 2,S
Seven-Machine Example Problem for Method 3 =mSmof,mzk~mo,~Z,,e Z0 2.0 0.2 1.~ 0.4
I l l l m U l a l ItIIUlI III
Machine ID 1 2 3 4 5 6 7 ComlnctN Row i.lne
7 3 6
Absolute S u m
o f "Negative" 9 6 3 5 2 2 6 Ml~ln# I~ 1 2 4 5
Sum of"Positive" 4 3 5 8 5 8 0 Somo,~/~azWSme~m,We 0-0 Q~ ~ 5
N~Aull l l n of N I g I ~ I ~ of PUlI~ 2.0 4.0 0.333 0.2
I,ilmmNla~ IIIUII IV
Conllncted Row Une
Constructed Row Une 7 5 3 6
7
Updated Absolute Sum of "Negative" and Som of "Posl~ve" Ml~c~cmI1~ 1 2 4
l~lt ~ I~,dm/NIIOlUll Scumof Mlolli~ 1~ 0 -
k~olm Sereol NmalkW~m o~I ~ v e 0.1167 - 0
/ /
NmmU~ I~lt V
ConCmc~l FlowUne
7 2 1 4 5 3 6
7 2 6
F~,ure 9
Updated .absolute Sum of "Nega1~e" and Sum of "Positive" Intermediate Results of the Seve~PMaehlne Example
IVla~hlne iD 1 3 4 5 Problem Using Method 4
Al~0/ute Sum of " I ~ " 6 0 5 1
Sum of "PosltNe" 0 5 2 5
Intem~llats I ~ u # III
machines are calculated. The steps of this method
Constructed Row Une are listed below. Figure 10 illustrates Method 5
7 2 1 3 6
using the seven-machine example problem.
Updated/~solute Sum of "Negative" and Sum of "Positive"
Mac~JlirleID 4 5
1. Among the machine pairs, select and connect a
pair of machines that has the largest amount of
~ ~ N
Row Une flow interactions to generate a semiconstructed
7 2 1 4 5 3 6 flow line. If more than one pair o f machines is
F/gur~S qualified, arbitrarily select one pair.
Intermediate Results of the Seven-Machine Example
Problem Using Method 3 2. Find a pair of candidate machines that has the
largest number o f flow interactions and does not
2. Figure 9 illustrates this method using the seven- belong to the same semiconstructed flow line. A
machine example problem. machine is a candidate if it is not connected to
any other machines, or if it is the first machine
4.1.5Method5 or the last machine of a semiconstructed flow
Method 5 can only be used for bidirectional flow line. If more than one pair of machines is quali-
lines. The objective of this method is to maximize fied, arbitrarily select one.
the total number of in-sequence movements in a 3. Connect the selected machines together. There
bidirectional flow line. Before applying Method 5, are three possibilities here. The first one is that
the total flow interactions between every pair of two machines are connected together to become
Journalof ManufacturingSystems
Vol. 17/No. 1
t998
MC M1 M2 M3 M4 M5 M6 M7
MC M1 M2 M3 M4 M5 ~ M7
M1 0 7 6 6 6 9 3
M2 0 7 3 3 6 5 M1 0 2 3 4 5 6 1
M3 0 6 5 8 2 M2 5 0 3 2 1 4 2
M4 0 5 4 7 M3 3 4 0 1 2 5 1
M5 0 2 8 M4 2 1 5 0 3 1 2
M6 0 3 M5 1 2 3 2 0 1 4
M7 0 M6 3 2 3 3 1 0 1
M7 2 3 1 5 4 2 0
I1~1
Sekct macl'dnes 1 and6. Itmatm I
Cnenect m ~ 1 anti mactline 6 togelher. 1 ~ flow from ma~tine 1 to machine6 is ~te grnetest.
Semi-Cor~ Row Line 1 : 1 - 6. Co~ect mlchine 1 to macNne 6.
Semi-Cor~ruct~ Row Line 1 : 1 - 6
II
Sekct mEl'mu 3 and 6. Itl~U
Cortnect machine 3 to machine6 of'b'lesemi-constructedflow line 1. The flow from m a ~ 2 to maclline 1, from maeNne 4 to maclllne 3, anti from mactline 7
Semi-ConslructedRow Une 1 : 1 - 6 - 3. to ma~ne 4 are the grmtest. We m~mrlly m~"t ~ 2 anti 1.
~ n e o t ma~ne 2 to m ~ l n e 1 of s e m i - ~ 0ow line 1.
It~ III Seroi~ Flow Line 1 : 2-1 - 6
Select m~ines 5 and 7.
Connect m a c ~ 5 and mactdne 7 togelher. It~ III
Semi-C~Rowl.kle 1 : 1- 6-3 The flow from macllkm 4 to machine 3, lind from machlne 7 to machine 4 are the ornetest.
S e r n i - C o n ~ Row Line 2 : 5 - 7 We arbitll~y ilekct n'BClklm 4 and 3.
Connect maclllne 4 to mactline 3 ~ buildanother8emi-conlRlctedflow line.
Itl~ IV Serni~ Flow LI~ 1 : 2 - 1 - 6
Select n ~ 2 aM 1. Semi-Con~ Flow Une 2 : 4 - 3
Connect m ~ 2 to machine 1 of the semi-conelxuctedflow line 1.
Seml-Conetr~ Row Line 1 : 2-1 - 6 - 3 Itm'atlonIV
Semi.Cone1~'uctedRow Line 2 : 5 - 7 The flow from machine7 to mscttine4 is l~e gre~est. Select maclYmes7 anti 4.
Connect m ~ 7 to machine 4 of s e m l - o ~ flow line 2.
It~V S e ~ Row Line 1 : 2-1 - 6
Select m ~ 4 and 7. Semi-Con~ Row Lh~ 2 : 7 - 4 - 3
Connet,'tmacftlne 4 to machine 7 of the Nmi-com'lxucted flow line 2.
Seml-Com'~tl~KIFIow Line 1 : 2-1 - 6 - 3 IWalbll V
Semi-Co--Row Line 2: 5 - 7- 4 The flow horn machim 3 to machine 2 and ~e flow h'om machine 5 to maclline 7 are the
Gr~tRCmacfikWearbib"adly sekct macl'dnm 3 and 2.
I t M t l N Vl 3 of serni-o0mlncted flow line 2 to maclYJne2 of semi-coneCuctedflow
Sekct ~ 3 aid 4. Une 1.
Connect m a c ~ 3 of semi-comtncted flow line 1 to machine7 of the s e m i - c o ~ Semi-~ Row Une I : 7 - 4 - 3 - 2 - 1 -6
flow line 2.
RnalRowLine: 2 - 1 - 6 - 3 - 4 - 7 - 5 ~ . a t m Vl
111eflow flora mclline 5 to mactline 7 is ~le g n ~ e ~
Connect mad11~ 5 to n'~hk~ 7 of semi-cot,ag'dCt~ flow I.
Figure 10 Rnel Row Line : 5 - 7 - 4 - 3 - 2 - I - 6
Intermediate Results of the Seven-Machine Example Problem
Using Method 5 Flare 11
Intermediate Results of the Seven-Machine Example Problem
Using Method 6
a new semiconstructed flow line. The second
one is that a machine is added to a semicon-
structed flow line. The last one is that two semi- the other is the largest and build a semicon-
constructed flow lines are connected to become structed flow line by placing the "from"
a larger semiconstructed flow line. machine in front of the "to" machine. If more
4. Repeat steps 2 and 3 until the entire flow line is than one pair of machines is qualified, arbitrari-
completed. ly select one pair.
. Find a pair of candidate machines that do not
4.1.6 Method 6 belong to the same semiconstructed flow line
Method 6 can only be used for unidirectional flow and the amount of flow from one to the other is
lines. The objective of this method is to maximize the the largest. A machine is a candidate if it is not
total number of in-sequence flow movements in a uni- connected to any other machines or it is the first
directional flow path. Method 6 is similar to Method or the last machine in a semiconstructed flow
5 except in Method 6 one has to satisfy the feasible line. Again, if more than one pair of machines is
flow direction constraint when connecting machines qualified, arbiU'arily select one. Since the flow
together. The steps of this method are listed below. line is unidirectional, for those machines that are
The seven-machine example problem is also used to the first ones of semiconstructed flow lines, one
illustrate Method 6 as shown in Figure 11. can only add machines in front o f them; that is,
they cannot be "from" machines. On the other
1. Among all machine pairs, select the pair of hand, for those machines residing at the end of
machines where the amount of flow from one to semiconstructed flow lines, they cannot be "to"
Journal of Manufacturing Systems
Vol. 17/No. 1
1998
10
Journal of Manufacturing Systems
Vol. 17/No. 1
1998
11
Journal of Manufacturing Systems
Vol. 17/No. 1
1998
M=9.
(a) (b)
5. Experiments 5.000
4.000 4.000
A large variety of experiments have been con- ~3.000
ducted to test the performance of the proposed lay-
j 3.000
2.000
1.000 1.000 1" l
out procedure in different problem scenarios when 0.000 0.000 I t i ) ) i i i J i"-
strategies are used. In Section 5.1, methods for eval- (C) (d)
5.1 Performance Evaluation now the optimal value of the criterion, which is also
Four performance evaluation methods will be obtained by enumeration. Figure 14 shows a flowchart
used to evaluate the solution quality and measure the describing the steps for calculating QI(Sj~,)*. First, as
performance of the proposed layout procedure. The shown in Figure 14, the proposed two-phase layout
first two methods measure the solution quality. The proc~ur is used to solve problemj with FLA method
third method measures the improvement in solution m and temperature reduction strategy n. As stated ear-
resulting from applying SA to the solutions found by lier in Section 4.2.1, Eq. (2) is used to evaluate the
FLA methods. The last method is the average CPU solutions found by SA at the second phase of the two-
time for solving a problem. Since the first three phase procedure. After the second phase of the two-
methods require some calculations, they are phase procedure, a solution, say Sj~,,, is found. One
described in detail as follows. then enumerates all the solutions of problem j. For
each criterion of problemj, one compares the value of
5.1.1 Absolute Measure of Solution Quality the criterion of every enumerated solution and finds
The first method is an absolute measure that mea- the optimal value of the criterion. After the best values
sures the proximity of a solution to the global opti- of all evaluation criteria of problemj have been found,
mal solution. For a problem j, let Sj,.,. be the solu- one uses these values as the new normalizing factors
tion found by the proposed procedure using FLA in Eq. (2) to calculate the normalized objective value
method m and temperature reduction strategy n, and of every enumerated solution. The solution with the
let S~ be the global optimal solution found by enu- best normalized objective value is the optimal solu-
meration. The solution quality of Sj,.,. relative to S~, tion. With these n^ormalizingfactors, the value of the
that is, QI(Sj..,.)*, is defined as follows: optimal solution, V(S), is not necessarily equal to one
because the normalizing factors do not necessarily
come from S~. Similarly, one uses these best values as
QI(Si'~")*= V(S;) (3) the normalizing factors to calculate the normalized
objective value of Sj~., which is ~Sj~..). After that,
^S
V(~,,,,,) and ~S~) are normalized objective values QI(Sj~,u,)* can be obtained using Eq. (3).
of Sj~,~ and S~, respectively. They are calculated using
a formula similar to Eq. (2), except now the normaliz- 5.1.2 Relative Measure of Solution Quality
ing factor for each criterion is no longer max,. or min*~. One disadvantage of the absolute measure is that
Instead, the normalizing factor for each criterion is it is very time consuming to calculate. This is espe-
12
Journal of Manufacturing Systems
Vol. 17/No. 1
1998
c~c.~te~(s~*= O(s~)^
1 Calculate0/lSx.~'= V---(~)V(S'~)
v(s~ F / p r e 15
Flowchart for Calculating ~e Relaliv Measure of Solution Quality
13
Journal of Manufacturing Systems
Vol. 17/No. I
1998
defined as the difference between the normalized because they are for minimizing the total backtracking
objective value of the initial solution found at the fin'st flow distance and minimizing the number of back-
phase and the normalized objective value of the solu- tracking movements, respectively. For each problem
tion after the improvement of SA at the second phase. scenario, different quantities of machines are possible.
Let Sj~ be the solution found by FLA method m for a In these experiments, for each problem scenario, 6, 7,
problemj after the f'n-stphase of the two-phase proce- 8, 14, 16, and 18-machine systems were considered,
dure, and let Sj~,n be the solution obtained after apply- resulting in six problem types. The 6, 7, and 8-
ing SA to Sj,,. with temperature reduction strategy n. If machine problems are referred to as limited-machine
absolute measure is used, then the improvement made problems, and the 14, 16, and 18-machine problems
by SA can be calculated as P(Sz~ ) - V(Si~). On the are referred to as extended-machine problems. For
other hand, if relative measure is used, then the each problem type, 100 randomly generated problems
improvement made by SA is IT(Sign)- IT'(Szm). were solved. As a result, there are totally 8400 prob-
lems in these experiments.
5.2 Experiment Design Six different FLA methods have been developed.
Five sets of experiments have been conducted. Because FLA Method 5 is only for bidirectional
For each experiment set, the performance of the two- flow lines and Method 6 is only for unidirectional
phase procedure with different combinations of FLA flow lines, there are effectively five possible FLA
methods and temperature reduction strategies was methods for each problem. Four possible tempera-
investigated for each of the relevant problem scenar- ture reduction strategies have been developed. As a
ios. Of particular interest was determining the effect result, there are 20 possible combinations of FLA
of the FLA method and temperature reduction strat- method and temperature reduction strategy. Each
egy on the solution quality relative to every applica- problem is solved 20 times using these combinations
ble evaluation criterion in different problem scenar- of FLA method and temperature reduction strategy.
ios. Also of interest was the improvement one can For every problem, the initial temperature is set at
make on the solution quality by applying SA to the 4.0 and the freezing temperature is set at 0.3. These
solutions found by FLA methods. temperatures are within their respective ranges as
The first four experiment sets test the perfor- suggested in Section 4.2.2. Although the chance for
mance of the FLA method and temperature reduc- SA to find the optimal solution will be higher for a
tion strategy relative to each of the evaluation crite- larger iteration limit, I,,~, there is no way of know-
ria listed in Table 1. Since these experiment sets are ing exactly what value 1re,o,should be in order for SA
only interested in each individual evaluation criteri- to find the optimal solution of a problem.
on, every problem tested within these experiment Furthermore, there are problems for which it is dif-
sets has a single-criterion objective function. The ficult or impossible to verify the optimality of their
fifth experiment set tests the performance of each solutions. In these experiments, ifI,,,~ is set too large
FLA method and temperature reduction strategy rel- then it is possible that every combination of FLA
ative to the multiple-criterion objective function. method and temperature reduction strategy will per-
Every problem tested within this experiment set has form so well that there is no way to tell which FLA
a normalized weighted multiple-criterion objective method or temperature reduction strategy has better
function with nonzero weights for all criteria that are performance. Therefore, in these experiments, for
applicable to the problem. each problem type, /max is set sufficiently large so
For the fLrSt, second, and fifth sets of experiments, that at least one combination of FLA method and
four problem scenarios are possible. This is because temperature reduction strategy produces satisfactory
the ftrst and second experiment sets are for minimiz- results. This is done by experimenting and testing
ing the total flow distance and maximizing the number some sample problems with different I , ~ values
of in-sequence movements, respectively. Furthermore, before the experiments. For 6, 7, 8, 14, 16, and 18-
the fifth experiment set has a multiple-criterion objec- machine problems, I,,~ is set to 100, 120, 150, 200,
tive function with nonzero weights for all the criteria 240, and 280, respectively. Since these experiments
applicable to each problem. For the third and fourth are not interested in the effects of the number of
experiment sets, only one problem scenario is possible temperature reductions, M, on the solution quality, it
14
Journal of Manufacturing Systems
Vol. 17/No. 1
1998
is arbitrarily decided that for every 10 iterations the first, second, third, and fourth sets of experi-
there is a temperature reduction. In other words, for merits, while Tables 14 and 15 summarize the results
6, 7, 8, 14, 16, and 18-machine problems, Mis set to of the fifth set of experiments. Tables 6, 8, 10, 12,
9, 11, 14, 19, 23, and 27, respectively. and 14 categorize the results based on the FLA
As for the fifth experiment set, the weights for method, and Tables 7, 9, 11, 13, and 15 categorize
minimizing the total flow distance criterion, maxi- the results based on the temperature reduction strat-
mizing the number of in-sequence movements crite- egy. Table 16 lists the average CPU time for solving
rion, minimizing the total backtracking flow dis- problems of different sizes. Finally, Table 17 shows
tance, and minimizing the number of backtracking the improvement in solution resulting from applying
movements criterion are arbitrarily set to 5, 3, 2, and SA to the solutions found by FLA methods.
1, respectively. The absolute performance measure Observations from the results of these experiments
only was used for limited-machine problems, while are summarized as follows.
the relative measure only was used for extended- From Table 6 one can see that, relative to the cri-
machine problems. The average CPU time and the terion of minimizing total flow distance, the solu-
average improvement resulting from the application tions of the limited-machine problems are all very
of SA to the FLA solutions will be collected from close to the optimal solutions. As for the extended-
the fifth experiment set. The experiments were con- machine problems, the average performance of
ducted on a SUN SPARC 10 workstation. every FLA method is close to one. This means that
for each of these extended-problems, the solution
5.3 Experimental Results found by every FLA method is very close to the best
The experimental results are summarized in solution of the 20 solutions found by using different
Tables 6--17. Tables 6-13 summarize the results of combinations of FLA method and temperature
Table6
Average Performance of Different FLA Methods Relative to the Criterion of l~linlmidngToTS]Flow Distance
0PM ffipcrforman~ measure)
FLA Method
Problem Type PM Method 1 Method 2 Method 3 Method 4 Method 5 Method 6
* denotes which FLA method has the best performance in every problem type.
15
Journal of Manufacturing Systems
Vol. 17/No. 1
1998
reduction strategy. However, it is observed that FLA as well. As a matter of fact, for both the limited-
Method 5 seems to dominate the other methods in machine problems and the extended-machine prob-
bidirectional problems. In addition, except for the lems, FLA Method 5 has the best performance in
seven-machine unidirectional loop flow path prob- every bidirectional problem. As for the average per-
lem and the eight-machine unidirectional nonloop formance of different temperature reduction strate-
flow path problem, the best performances in the rest gies relative to the criterion of maximizing the num-
of unidirectional problems are shared between FLA ber of in-sequence movements, one can see from
Methods 4 and 6. As for the average performance of Table 9 that although the third temperature reduction
different temperature reduction strategies relative to strategy wins the majority of the best performance,
the criterion of minimizing total flow distance, it can all temperature reduction strategies have very close
be observed from Table 7 that no temperature reduc- performance.
tion strategy significantly dominates the other As for the performance of the FLA method rela-
strategies; however, the third strategy seems to have tive to the criterion of minimizing total backtracking
slightly better performance. flow distance, Table 10 shows that FLA Method 1
Relative to the criterion of maximizing the num- has the best performance in every limited and
ber of in-sequence movements, one can see from extended problem. In addition, for the performance
Table 8 that all the solutions of limited-machine of the temperature reduction strategy relative to the
problems are extremely close to the optimal solu- criterion of minimizing total backtracking flow dis-
tions. As for the extended-machine problems in tahoe, Table 11 shows that all strategies have very
Table 8, one can see that only FLA Methods 5 and 6 close performance with the third strategy, winning
perform well--their average performance values are half of the best performances. From Table 12, one
close to one. The other FLA methods do not perform can see that all FLA methods have very close per-
7able7
AveragePerformance of Different T m p e n t m Reduction S t r t t q ~ lblttive to the Criterion of Minimi~ng Total Flew Distance
O'M". perromnce m~ure)
TemneratureReductionStrategy
ProblemType PM Temp 1 T~'mp2 Temp3- Temp4
6 machines, unidirect., nordoop .985342 .985568* .985436 .984679
6 machines, unidirect., loop .995942* .995518 .995413 .995113
7 machines, unidirect., nonloop a .970075 .970591 .971543 .971649*
7 machines, unidireet., loop b .986467 .986231 .986644* .986026
8 machines, unidirect., nonloop s .962736 .964257* .963472 .961099
8 machines, unidirect., loop o .977568 .978254 .979226* .977833
6 machines, bidirect., nonloop 1 .984495 .985155* .984811 .983834
6 machines, bidirect., loop u .998252 .998161 .998115 .998612*
7 machines, bidirect., nonloop t .971808 .971195 .970651 .973551*
7 machines, bidireet,, loop e .990693 .990736 .991088* .990048
8 machines, bidirect., nonloop .963843 .964041 .965206* .963081
8 machines, bidirect., loop .983973 .983544 .984762* .983534
14 machines, unidirect., nonloop .969886* .969446 .969832 .968572
14 machines, unidirect., loop .981681* .981356 .981366 .980809
16 machines, unidirect., nonloop r .971869 .971383 .972569* .971295
16 machines, unidirect., loop e .982732 .982906* .982739 .982879
18 machines, unidirect., nonloop 1 .974601" .974256 .973668 .974211
18 machines, unidirect., loop a .983340 .983661* .983311 .982849
14 machines, bidirect., nonloop t .969608* .969344 .969361 .968716
14 machines, bidirect., loop i .980637 .980825 .980761 .981578*
16 machines, bidirect., nonloop v .973326 .971439 .973989* .972552
16 machines, bidirect., loop e .982680 .982403 .982730 .982843*
18 machines, bidirect., nonloop .974163 .974410 .975139" .973769
18 machines, bidirect., loop .982914 .983492* .982886 .983089
* denotes which temperature reduction strategy has the best performance in every problem type.
16
Journal of ManufacturingSystems
Vol. 17/No. 1
1998
formance relative to the criterion of minimizing the egy. With this knowledge, it becomes easier to select
number of backtracking movements, with FLA appropriate FLA methods and temperature reduc-
Method 4 winning all the best performances. Table tion strategies for different problems; as a result,
13 shows that all temperature reduction strategies good solutions can be found more effectively and
perform equally well, with the second strategy win- efficiently. This is very helpful especially when
ning half of the best performances. there are constraints of time and other computation-
From the discussion above, one can understand al resources. For example, in the fii~h set of experi-
the performance of different FLA methods and tem- ments, the weights assigned to the criteria are 5
perature reduction strategies relative to every applic- (minimizing total flow distance), 3 (maximizing the
able criterion in every problem scenario. The results number of in-sequence movements), 2 (minimizing
of the experiments indicate that the selection of FLA total backtracking flow distance), and 1 (minimizing
method can be based on the evaluation criteria in the the number of backtracking movements). Larger
objective function of a problem. This is because one weights are assigned to the first two criteria. With
FLA method seems to dominate the other FLA this assignment of weights and the knowledge
methods for all problems in the experiments for a learned from the first four experiment sets, it is
given evaluation criterion, with the exception of uni- expected that FLA Method 5 will have the best per-
directional problems for the evaluation criterion of formance in bidirectional problems and FLA
minimizing the total flow distance. Although one Method 6 will have the best performance in unidi-
does not see such significant dominance among rectional problems. It is also expected that the third
temperature reduction strategies, the results from temperature reduction strategy should be able to per-
Tables 7, 9, 11, and 13 more or less provide some form well. In other words, to solve this problem, one
ideas on the selection of temperature reduction strat- should consider using FLA Method 5 (for bidirec-
Table8
Average Performance of Different FLA Methods Relative to the Critet~n of MaYimi~iqgthe Number of In-Sequence Movements
(I'M = performaaee measure)
FLA Method
Problem Type PM Method 1 Method 2 Method 3 Method 4 Method 5 Method 6
* denotes which FLA method has the best performance in every problem type.
17
Journal of Manufacturing Systems
Vol. 17/No. 1
1998
tional problems) or FLA Method 6 (for unidirec- reduction strategy has the best performance in the
tional problems) and the third temperature reduction majority of problems. Table 16 lists the average CPU
strategy first before other methods and strategies. time for solving a problem of different sizes in the
The results of the fifth set of experiments shown filth experiment set. As seen in Table 16, the pro-
in Tables 14 and 15 prove these speculations to be posed two-phase procedure is very effective in terms
true. From Tables 14 and 15, one can see that FLA of CPU time. In addition, Table 17 lists the average
Methods 5 and 6 have the best performance in bidi- improvement made by SA for each problem type in
rectional problems and unidirectional problems, the fit~la set of experiments. Table 17 shows that the
respectively. In addition, the third temperature improvement is significant for some problems.
Table9
Average Performance of Different Temperature Reduction Strategies Relative to the Criterion of Maximizing the Number
of In-Sequence Movements (PM = performance measure)
* denotes which temperature reduction strategy has the best performance in every problem type.
Table 10
Average Performance of Different FLA Methods Relative to the Criterion of Minimizing Total Backtracking Flow Distance
0PM = performance measure, abs = absolute, and tel = relative)
FLA Method
ProblemType PM Methodl Method2 Method3 Method4 Method6
* denotes which FLA method has the best performance in every problem type.
18
Journal of Manufacturing Systems
Vol. 17/No. 1
1998
Because of the relatively small mount of CPU time paths. In addition, linear single-row flow path con-
for solving one problem, it is concluded that the figurations are popular in practice because of their
improvement made by SA is worth the effort. simple structures and relatively easy flow control
problems. As pointed out earlier in the paper, tradi-
6. Summary and Conclusion tional machine layout procedures ignore flow path
This paper proposed a two-phase layout proce- characteristics and almost all of them restrict the
dure to solve the machine layout problem with a lin- flow path to a straight line. Unlike traditional layout
ear single-row flow path in an automated manufac- methods, the proposed layout procedure considers
turing system. Many modern automated material the characteristics of the flow path and does not have
handling devices transport material on fixed flow the same restriction. Because of the consideration of
Table 11
Average Performance of Different Temperature Reduction Strategies Relative to the Criterion of MilaimizingTotal Backtracking Flow Distance
(PM = performance measure, abs ffi absolute, and tel ffi relative)
* denotes which temperature reduction strategy has the best performance in every problem type.
Table 12
Average Performance of Different FLA Methods Relative to the Criterion of Minimizing the Number of Backtracking Movements
(PM ffi performancemeasure, abs ffi absolute, and tel ffi relative)
FLA Method
Problem Type PM Method 1 Method 2 Method 3 Method 4 Method 6
* denotes which FLA method has the best performance in every problem type.
Table 13
Average Performance of Different Temperature Reduction Strategies Relative to the Criterion of Minimi~dng the Number of Backtracking
Movements (TM = performance measure, abs ffi absolute, and rel = relative)
* denotes which temperature reduction strategy has the best performance in every problem type.
19
Journal of Manufacturing Systems
Vot. 17/No. 1
1998
flow path characteristics, appropriate evaluation cri- ear single-row flow path, it is the first layout proce-
teria can be selected for different layout problems. dure that has been tested for the performance of dif-
Four evaluation criteria have been introduced to ferent FLA methods and temperature reduction
assist in evaluating the layout design for different strategies relative to every applicable evaluation cri-
problem scenarios. Minimum safe distances teflon in every problem scenario. Many experiments
between machines were introduced for the safe oper- have been conducted. The results of these experi-
ation of machines, storage space for WIP, move- ments provide vital information on selecting appro-
ments of personnel, and many other purposes. priate FLA methods and temperature reduction
Suggestions were also made to prevent the overlap- strategies for solving different problems.
ping of machines. The proposed layout procedure From the experimental results, it was also observed
combines FLA method and SA. Two new FLA that the proposed solution procedure performs
methods have been proposed. Four of Hollier's extremely well for the limited-machine problems. For
methods ~2have been adopted and modified so they the extended-machine problems, although their solu-
can be applied to problems with duplicate machines. tions are not compared with their respective optimal
A normalized multiple-criterion objective function solutions, it is believed that near-optimal solutions will
has been proposed to incorporate different evalua- be obtained for the extended-machine problems if the
tion cflteria together so that the layout design can be number of iterations, I,~, is large enough. In fact, if
evaluated. Initial and freezing temperatures setting the time constraint allows, one can further improve the
strategies and four temperature reduction strategies chance of achieving an even better solution of a prob-
for SA have also been suggested. Although the pro- lem by any one or combination of the following
posed layout procedure is not the first method that means: using the solutions obtained from other FLA
applies SA to the machine layout problem with a lin- methods as initial solutions, increasing the iteration
Table 14
Average Performance of Different FLA Methods Relative to the Normalized Weighted Multiple Criteria in the Fifth Experiment Set
(PM = performance messure)
FLA Me~od
Problem Type PM Method 1 Method 2 Method 3 Method 4 Method 5 Method 6
* denotes which FLA method has the best performance in every problem type.
20
Journal of Manufacturing Systems
Vol. 17/No. 1
1998
Table 15
Average Performance of Different Temperature Redaction Strategies Relative to the Normalized Weighted Multiple Criteria in the
Experiment Set (PM = performance measure)
* denotes which temperature reduction strategy has the best performance in every problem type.
Table 16
Average CPU Time for Problems with Different Quantities of Machines
Quantity of Machines 6 7 8 14 16 18
Average CPU time (seconds/problem) .015538 .021125 .026613 .084925 .119688 .16855
Table l 7
Average Improvement of Every Problem X~)~e
Quantity of Machines bidireet., nonloop unidireet., nonloop bidireet., loop unidireet., loop
21
Journal of Manufacturing Systems
Vol. 17/No. 1
1998
22