Sie sind auf Seite 1von 19

This article was downloaded by: [Universitaets und Landesbibliothek]

On: 05 September 2013, At: 06:44


Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK

International Journal of Production Research


Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/tprs20

Numerical taxonomy applied to group technology and


plant layout
a
A. S. CARRIE
a
Department of Production Engineering, University of Strathclyde, Glasgow, Gl 1XJ, Scotland
Published online: 27 Mar 2007.

To cite this article: A. S. CARRIE (1973) Numerical taxonomy applied to group technology and plant layout, International
Journal of Production Research, 11:4, 399-416, DOI: 10.1080/00207547308929988

To link to this article: http://dx.doi.org/10.1080/00207547308929988

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the
Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
INT. J. l'HOD. RES., 1973, VOL. 11, No.4, 399-416

Numerical taxonomy applied to group technology and


plant layout

A. S. CARltLE*

This paper identifies tho need for a technique which can assess whether a functional
or a group cell system of production is most appropriate in a specific case as well as
develop uhe appropriate system. The paper describes the technique of numer-ical
taxonomy and shows how it may be applied to both group technology und plant
layout. A computer programme for production flow analysis is presented. The
Downloaded by [Universitaets und Landesbibliothek] at 06:44 05 September 2013

ccuuponont-mechinc matrix method of production flow analysis. as presented by


Burbidge, is shown to be of limited value in practical cases and an improved method
is proposed. The use of numerical taxonorny in developing functional-type layout
is also described.

Introduction
It is an acceptod principle that the design of a production system should
take account of the work flow structure of the parts to be made (H ollier and
Corlett 1966, Holstein and Berry 1970, Carrie 1971 a). The analysis of the
work flow structure in batch production can involve a considerable data
processing problem. Many pieces of information, such as batch quantity,
batch frequency, number of operations, sequence of operations, operation
times and set-up times, will be required for each component. The use of a
computer is almost esscntial in all but the smallest practical problems.
Most. published computer programs, however, analyse work flow with the
objective of designing a specific type of production system, usually either a
functional layout or It multi-product line or group layout (Hollier 19G;1, Buffa
et al. I !)(i4, Lee and Moore 1907, Burbidge 1971). What is gcnerally lacking
is a simple and efficient technique which is capable of showing which type of
layout is most suited to a particular case. It should also develop the specifica-
tion of the system either up to the stage where manual intervention in the form
of subjective judgement becomes necessary, or alternatively, to provide input
data for an existing programmc, The former alternative is, of course, prefer-
able, since existing programmes tend to have their individual advantages and
disadvantnges (Carrie 1971 b).
The author's interest in numerical taxonomy arose because the technique
appeared to have the necessary flexibility to fulfil these requirements and
could be readily implemented on a computer. This paper attempts to show
the extent of thc usefulness of the technique.

Numerical taxonomy
Taxonomy is the science of biological classification of objects based on their
possession or lack of relevant charucteristics. When the extent to which they
.. Depar-tment of Production Engineering, University of Strathclyde, Glasgow a 1 lXJ,
Scotland.
Presonteo at tho 2nd International Conference on Production Hosearch (Copenhagen) August 1973.

Published by Taylor & Francis Ltd., IG-14 Macklin Street, London WC2B 5NF.
400 A. S. Carrie

possess each of these characteristics can be expressed numerically the objects


can be classified by numerical taxonomy (Sokal and Sneath 1963). Numerical
taxonomy provides algorithms for the study of similarities between objects
in a quantative manner, contrasting with the classification techniques of group
technology which tend to be descriptive.
Numerical taxonomy is a method of analysis rather than ,t formula to be
executed. It involves three stages.
(I) Prepare It data matrix. This indicates which characteristics are either
present or absent, for example, whether a component is rotational or
non-rotational and whether or not it possesses a through-going central
hole, and can be indicated by a series of binary digits rather like an
Downloaded by [Universitaets und Landesbibliothek] at 06:44 05 September 2013

extended code. Other characteristics must be specified by numerical


values such as the diameter of a hole, the batch quantity or the weight.
(2) Compute ,t similarity eoeffieient matrix. From the information con-
tained in the data matrix the similarity between each pair of objects
ean be evaluated. This similarity is based on the extent to which the
objects share characteristics. The coefficient is normally defined to
have a maximum value of 1·0 when two objects have identical charac-
teristics, and a minimum of 0·0 when they have none in common.
(:1) Perform cluster analysis. Cluster analysis examines the similarity
bctween each pair of objeets and forms groups of objects so that within
eaeh group the objects are highly similar to each other. An analogy is
to think in terms of the distance between objects in space. If each
object cn.n possess up to J\{ characteristics they can be- considered as
occupying positions in an lVI-dimensional space. Objects whieh are
highly similar to one another are close together in this space. Cluster
analysis seeks to find groups of objects forming distinct clusters.
Numerical taxonomy is a teehnique of eonsiderable flexibility. No single
formula exists for defining the similarity between objects. The similarity
coeffieients nmy be calculated by the method most appropriate to the parti-
cular problem. Similarly, many ulgorithms have becn developed for cluster
analysis. Thcy present their results in a variety of different ways, and may
also be selected to suit thc case being studied.

Application to group technology


The choice between the two types of layout, functional and group, can
normally be made by determining whether or not distinct families of com-
ponents can be found among the parts to be made. If such families exist and
provide viable loads on the corresponding machine tools, the group layout
principle can bc adopted, but if not the functional approach must be resorted
to. A technique for the design of production systems must therefore provide
facilities for examining this equation.

A pplicrrtion to prodmction. [lou: nnnlysis


Burbidge (I !J71) advocates the use of production flow analysis for the forma-
tion of groups of machines and families of components for group technology.
This technique is concerned with the analysis of operation sequences and work
Numerical taxonomy applied to group technology and plant layout 401

loads, that is the data identified in the introduction to this paper. It is more
suitable for computer implementation than the alternative of using group
technology coding techniques. These involve a considerable manual effort in
preparing the code numbers, which may vary considerably in significance and
usefulness from one company to another. The central part of Production
Flow Analysis, group analysis, utilizes a machine-component matrix indicating
which machines are required for the production of each component. The
matrix will initially appear as a random pattern, but by sorting the matrix and
rearranging the sequence in which the rows and columns are printed it is

Fig. 1
Downloaded by [Universitaets und Landesbibliothek] at 06:44 05 September 2013

..
!i
PART HUHBER'

"
~
&: 1 2 )
• J II , ".. ., d III .1 ., ., ".. 20 I' Uti ZtU "If ra" ~ 3' JIB Jt Sf
· .- " 5 .- .- .-

· ·· · · · · · " · .-
K. . .
b
.- v v v
.-
C v "

""·
v v
v
"

v
" " " " "
•, · · ·
· " "
v •
" .- ·
"
"
,"
"
·· "
· ·
v
"

"
I " "
·
• 0 "

v
•• • " v v
"

"
" ·
" v
" .- v "
"· · ·· •
v

'"
"0

· "" " · "" " " • "
"
" v
"
v v

.
/I
" " ""
· " " " · · .- " " "
"
• "
"
·" "
• ··
v

~
T • "
" · ·
Production flow analysis-original component-machine matrix, from Burbidge (1969).

Fig. 2

...z: PART HU/'1B~RS

i
~ '1 ., Zt 2131 7 '0 f8 315 is fI 2023 2S2,1~ 6 , If If 1"5 30 '2 33 8 (1/ (ftl
"" 51
· ·· ··o' ·· ·· ·· ·· · ·
Z
b
d.
·
. · ·· · ·· · · · · ·
·
",. ., .-
R

c
···· · ·· •• ·· ·· · · · · · · v

h
II ··· ··· ··· ··· ··· ··· ··· ·· ·· •
•, Fa ·· ·· ·· ·· ·· · · · · ·
··· ··· ··· ··· ··· ··· ··· ··· ·· ·
0
p


···· ··· ··· ··· ·· .. ..
T
• ·
· ·· · •· ·
I
I

Production flow analysis-sorted component-machine matrix, from Burbidge (1969).


['J.P.R. 2 D
402 A. S. Carrie

possible to find groups and families. Eigures I and 2, from Burbidge (1969),
illustrute the principle. Notice the distinct pattern which has emerged in fig. 2.
An algorithm which rearranges the sequence in which the items to be classi-
fied arc written so that highly similar items are brought together has been
published by Ross (1969). Since this is exactly what production flow analysis
requires, its suitability was examined. This examination was carried out
initially by MacAuley (1970, 1972) who concluded that the technique' appears
to gi vc satisfactory results " and recom mended that it 'be tested on a real
problem '. Subsequently the author has used the technique in several real
problems as one of several facilities in a suite of programmes being developed
to dctcrminc thc cxtcnt to which the work of the design of a production system
mlLY be done by computer.
Downloaded by [Universitaets und Landesbibliothek] at 06:44 05 September 2013

The result of applying the technique to the exampic quoted by Burbidge is


shown in fig. a. 'I'he matrix, printed directly by a computer, indicates identical
groups and familics to those of Burbidge.

A sub-routine for group analysis


The appendix to this papcr provides a listing and description of a sub-
. routine for group analysis, sub-routine GRPANALYSIS. It contains a large
number of commcnt statements which clearly indicate the sequence of opera-
tions curried out by the sub-routine. A few additional comments may be
appropriate herc. By means of calls to other sub-routines GRPANALYSIS
applied the three stages of numerical taxonomy twice; firstly to classify the
components and secondly to classify the machines. GRPANALYSIS assumes
that a data file giving details of the operation sequences of the component
exists on disc store, Sub-routines BETCOM and BETl\iAC interrogate this
filc and create machine-component data matrices on disc file. BETCOM
creates a matrix with components as rows, machines as columns, whereas
BETMAC creates one with machines as rows, components as columns. Sub-
routine SCOEF I. culoulates the similarity matrix by examining the data
matrix created by BETCOM or BETMAC. Cluster analysis is carried out by
sub-routines PJ'tll\iTltEE and SLCA which are based on Ross's algorithm,
which was modified slightly to allow the use of disc backing store and to suit
the software and mcthod of implementing the programmes on the particular
computer. MATPl~INT prints the machine-component matrix in the form
shown in figs. 3, 4, 5. As input parameters sub-routine GRPANALYSIS is
provided with details of the data file and a list of the components to be
cxamined. It is also supplied with a code for each of the machines or other
facilities which will be encountered in the data file specifying how that machine
is to be considered in constructing the machine-component matrices. This
code number contained in the array MA, instructs the computer which of three
courses is to be adopted. The machine may be ignored, for example in the case
of a manual operation or inspection facility, it may be treated as an individual
machine, or it may be grouped together with other machines. This enables the
processing requirements of the components to be examined in varying level of
detail, for example, by considering all machining processes together, or by
considering all turning processes together or all capstan lathes, or all capstan
lathes of an equivalent capability or each machine individually. Table I,
showing part of a plant list used in one study, illustrates this principle. The
Fig. 3 Fig. 4 Fig. 5

,,
,
~tGMgEFjITKP'Ot&NRO"
1~·~~;~~~.1~.··a-···1'
,,"'3 '-~-~:.--~'~-- .. --N', ,.- ....
.... ... · . ,,, "
, •••••••• ·, ••••• - . · · , · · · · . ·•• 1

• • 31 1

..... . ·• •• ,,,
...·.
1
~. "

j-!....~.-!- ."
,,.
117

,, , ,.."
S,
~

...
_] _1-!~....!..!.!...

••• _-• ••
-,-.---r 2 •• * • 1 I

,
.., ·· ....• .....· ,, u'
IS' g

,,
1 _*!!.* '30

......
uo
,,. ••••
•••• - ''''15
Downloaded by [Universitaets und Landesbibliothek] at 06:44 05 September 2013

15 - 12 1 >,7 ;:::
, 1 1___
1

.,'"'""
,
17 ,o~ EO'
"
lO ••• lO 5 ,~" ..~
ec ,

.
__1.~ ~

,., ••••
~1_

,, "'. , -.... ,, ,."


-~- .

, "'.
~-
-_. -~-

2]~'*-" ~ '~2" ~. , hO
~ --~~ ~

-.~=--- , '8'
H 2S • 8 17

,. · · . . ·
~

;;;-
, ,
----
,
,,, • • ..... ......
7. I
29 IF
--
* 0 >, •
, .. , 26' !-l
Z
12
,
- ..
.....,,,
.....
,
,
····~1 ~ ,{
..2
"'- 15• , .,
,,~

·I...
.. ....... · ,
,05
62
c
;:::
<:>
,3
,0 ,,
24
,, ,0 ,, ***. ,3
24 17• , ".
HO , ·
I
I
.-
". ......... • . ·•. ,, !S •

;.l
""~
• ,,
,.,•
., , .,
,.,•

••• • 6 \2.

,,, • ...••. ,, , ., 82
H ••• ',8

, • .*.*• ,,
,
3'
- -•
31 13 ,~.
.~ I

.... · · · , • ,
, U· ~ ...
,, ...
I
1 7 .,..
l7
,, ••
22 ,
.
27 14
,2< I

... ·.. .. ·· ,, l,'l"


..
\2'
'"
,.,, ..••••• H' 1 1t1Z
8
, 1 8 s
,, ....... ·
;19

. ··
1

.,,,
,30

.
,I.

., ,,
~.*

• ,
""-e<:>
,,, ,6 ..
•••• ,- ra u
,~

•••*. 10 >fu
,e
·
I

.
,, ··....
••••
·
H 20
H >§
, ,,, , •
e • 11
2l ••• 22
...
.. .. ... • ,, ",., ,
.
.,2

,, •• •• ,
••• \1' ~

H. , ...
'13
)4
•• 1 34 18 1 '"o

,,, I
,
,07
,, I r0 7
I

•• *. • ~
.~ •• "! 1 - --I. I 460
21
*!'- •• . • ,, ,,0 c
...... ,,, " , ••• *'1'/
6 83
, -,- , •·
1
-9'
••••• ...
. '0' .8
9 1 (,~
J
". , ·. · ·
I

,,
He
11
2'
. t.··... , 23 ,*. •• •
••••• 2' 24 I

,2' ,,.. • •
,,~ .. · ,, 1 U•
HZ ""s,.,;:::
3~
l8
3l
L .
t .
• ..
.-:
.1
•• •
30
28 10 1.
,
,
,.3
lob
,,, ··.....
...
. ·
,, ,
I 22"
26'
-e
,,, , • ·....
3l
· ,,
!ft ••• '00
35._'-
H , I 1

•.. •
3S
•• ,
4 1
,.7
~
r0 7 ;;>'
;:::
· ·
I- 1

·.• ...... · ,, "',.,


])
.. •..
, • .,
~

,~·_· ~~.1~~· ~··'1


.',,
13'
A~GHgef J I T~PSUL 8N I!.PH __ 11 1 .. '33 ;;>'
:i ze
· .. ,,
.. ,
16 2t·v

., .... ...• ,..... ..,........


3
26· ""c
;::

..
I ~

20 1 I ~~ ~,

*
N i ,---.:.-.~,~._ -~.,

II
c
N - I -- il'o-
0
Component-machine matrix for the Component-machine matrix in a valve Component-machine matrix in a computer w
same problem produced by computer
algorithm.
manufacturing study. I manufacturing study.
404 A. S. Carrie

Table l. Plant list printout (showing machine categories for machine component
matrices)
PLANT LIST
MACHINE cOMPUTeR 11TH PROceSS SIMILAR
, syMBOL cODe N~.
20/1 73
COPE Type
31f 1
MACHI~ES
1
,
,,,
2 20/~ n 314 1
3 20/7 S8 32.) 2
4 1.019 S5 32J 2
5
6
20/1 0
20/"
72
rr
32.1
33,) ,, 2
3
7 20H 78 330
37.) , 3
8
0
20/0
05/.5
71
tl8 411 ~ 5"
Downloaded by [Universitaets und Landesbibliothek] at 06:44 05 September 2013

,,
10 08/~
08/)
0811
15S
158
153
51.)
512
5'2
J
j
.5
6
7
7
12
08/" 157 52,1 j 8
,,,
'3
08/.5 156 53,) .5 9
15 , 01 ~ 131 543 .5 10
16 '011 130 S4:l j 10
17 '0/) '34 54:1 .5 10
18 ,0/.5 132 54" j )0
10 ,0/" 133 S4S .5 10
20
2'
n911
noO
SO
52
617
617
~
~
,1
,1
,,
21. 00 I ~ ~, 617 ~

B 12/1 122 721' ) ,2


24 '2/~ 123 721 ) Il
25 121 s 124 72i' ) ,2

first column of the table shows the column number of the component-machine
matrix when machines are treated individually. The last column specifies the
matrix column when machines of similar capacity are grouped together. The
first two machines are light bench and pillar drills, the next three are multi-
spindle drills, and so on. The fifth column indicates the process type, with the
first eight machines being drilling machines, the ninth a gear hobber, the next
ten are grinding machines. The fourth column gives the British Standard
Machine Tool Classification'].

Definition of similarity between pairs of components or machines'


In the example problem the similarity coefficients between pairs of com-
ponents were defined as :
S .. = n(Ri_R j )
'! n(Ri-R j )

where Sij is the similarity coefficient between components i and j and R, and
R j are the sets of machines required to process components i andj respectively.
That is, the similarity between components is the ratio of the number of
machines found in the row of the component-machine matrix created by
BETCOM belonging to component i and also found in the row for componentj,
to the total number found in either row. In the example problem, component
22 requires processing by machines F, J and I while component 16 requires
E, F and J. The number appearing in both rows is 2, (F and J), and the

t British Standards Institution, Classification of Metalworking Machine Tools by


Types, BS 4640 : 1970.
Numerical taxonomy applied to group technology and plant layout 405

number is either in 4 (E, F, I and J) so that the similarity between the two
components is 0'50. The same definition was applied to the machine-
component matrix created by BETMAC to obtain the similarities between
machines. Thus the similarity between machines A and G in the example
problem is 0'50, i.e. 4/8.
SCOEF'1 also allows two other methods of defining the similarity coefficients.
These provide a slightly different emphasis in cluster analysis which may be
helpful in some cases.

The machine-component matrix in practical problems


The technique described above, having proved highly satisfactory in the
Downloaded by [Universitaets und Landesbibliothek] at 06:44 05 September 2013

hypothetical problem put forward by Burbidge, has been applied to several real
problems in industry. Among these were projects in a valve manufacturer
and in a computer manufacturer. The study in the valve manufacturer was
concerned with the company's 'standard products '-a range of butterfly
valves which accounted for most of the output. The operation sequences of
the components making up these valves were examined and the components
arranged in twenty-four packs so that within each pack all components had the
same operation sequence. Eighteen machines or other processes were involved
in the manufacture of the components. Figure 4 shows the matrix derived
by the group analysis programmes. The rows represent component packs,
the column the machines. Three distinct families have emerged, packs 1 to
17,6 to 21 and a to 20, but this' successful' result is qualified by three factors:

(1) there are rather more' exceptions' (asterisks outside the dense blocks)
than Burbidge appeared to anticipate;
(2) some machines have to be duplicated between groups;
(3) some packs (for example 1 and 6) could equally well be assigned to
either of two families.

In the computer manufacturer the parts studied were components of the


frames in which the electronic assemblies of the various units such as line
printers and disc drives were housed. Over three hundred components were
analysed and sorted initially into 46 packs with similar machining processes.
28 machines or processes were required for their production. Figure 5 shows
the derived matrix, in which the rows represent the component packs. Each
row number is the number of a typical component in each pack. Three families
have emerged, packs 31 to 264, 195 to 166 and 206 to B, The same three
factors qualify the' success' of this result, but to a much greater extent than
in fig. 4. The exceptions appear to outnumber the straightforward packs
although replanning could certainly eliminate some of them. The machines
not required in more than one group are a minority. Several packs, 83 to 229,
could be assigned to either of the first two families while 26:3 should be assigned
to the first family.
As a result of these factors the boundaries of the groups and families are
much less distinct than in Burbidge's hypothetical example. Figure 5 is, in
fact, typical of the matrices obtained in the real problems analysed by this
method. Nor have these real problems been taken from a limited area of the
metal cutting field; they have covered the full range of processes found in the
400 A. S. Carrie

engineering industry and extended beyond this field, for example into plastic
moulding manufacture. Three questions arise from this work:

( I) Is Ross's algorithm acceptable as a means of sorting the machine-


component matrix 1 If the answer to this question is yes,.then other
questions must be answered.
(2) Can families and groups suitable for group production be formed from
thc matrix 1 If thc answer to this question is no, then:
(H) Is production flow analysis itself a viable technique 1

Tho answer to the first question undoubtedly is yes. Figure 3 shows that
Downloaded by [Universitaets und Landesbibliothek] at 06:44 05 September 2013

Ross's algorithm is an effective means of sorting the matrix. In certain points


of detail its result may not be optimal, for example the position of pack 263 in
fig. 5. 'I'hese arise due to the algorithm's method of constructing groups.
High similurity relationships are considered first, items with lower similarity
are added '1t the end of groups, so that in fig. 5 a group consisting of packs 3]
to 22!l had been formed in which all the relationships were at least as strong as
that of 20;{ to any of these packs, so that 263 was added at the end of the group.
Tho procedure possibly could be refined, but it is unlikely that the results would
justify the increased computer time. Experiments at the University of
Strathclyde have shown that it is unlikely that any other approach would prove
either more effective or economical. They also show that Burbidge's assertion
that manual sorting of the matrix is suitable for problems with up to 2000
components is grossly over-optimistic.
The three factors mentioned above, which limit the usefulness of the matrix,
are thc result of practical facts of life not the result of the algorithm used. They
cast serious doubt on the use of the matrix for forming families and groups.
Figures ;{, 4 and .5 show that the matrix technique is certainly capable of
indicating in It visual form whether or not families exist. The second question
arises whether these families of components and the groups of machines
necessary for their production can be specified. The answer to this question
is undoubtedly no, for one could not form any machine groups from fig. 5
without studying the work loads produced on the machines by the various
component familics one could form tentatively from the matrix. The matrix
givcs no indication of work loads. In his paper Burbidgc stated that it would
bc necessary to check work loads only on those machines that are required in
more than one group, presumably on the assumption that these would be few
in number. If thcse machines were few in number matrix sorting would be
satisfnctory. Sincc this generally appears not to be the case, the usefulness of
the mutrix is rcstrictcd to providing a visual impression of relationships
between components which may be useful in interpreting the results of
othcr methods, or in providing tentative groupings for evaluation by other
methods.
Some writers (Edwards] !l72) have rejected production flow analysis as
being' a good idea which fails in practice'. Nevertheless, the failure of a detail
of technique, matrix sorting as a means of forming families, does not necessarily
invalidate the principle. The author believes that the answer to the third and
final qucstion posed above is yes, for an alternative to the matrix technique
may be developed to yield a satisfactory method of family formation.
Numerical taxonomy applied to group technology and plant layout 407

Family formation by successive analysis of similarity


In cluster analysis a level of similarity coefficient could be selected which is
considered to be the minimum level of similarity between components at which
they may be grouped together for family production. If components with less
than this level of similarity were to be grouped together there might be too great
a diversity of components for group production methods to be effective. In
addition, a minimum on the number of members in a family could be specified
to prevent the formation and analysis of a large number of ~ery small groups of
components.
These two principles form the basis of an algorithm by Wishart (1969).
When applied to the hypothetical problem (fig. 3) with the parameter values
Downloaded by [Universitaets und Landesbibliothek] at 06:44 05 September 2013

indicated the result shown in table 2 would be obtained. This form of output
shows the composition of the families with no ambiguity. It lists the' excep-
tions ' separately, so that they can be examined and their production methods
replanned if this were considered desirable.

Table 2. Groups and unassigned items for specified similarity level and minimum
group size

Groups at similarity level=0·800


Minimum cluster size =3
Group Quantity Members
1 6 1 3 5 15 17 20
2 4 2 12 13 24
3 4 8 14 19 26
4 6 4 6 9 11 21 30
Unassigned items
7 10 16 18 22 23 25 27 28 29 31 32 33 34 35

Wishart's algorithm commences by setting the similarity level initially to


1'0, so that groups of identical components are found, and then gradually
reduces the level until all 'components have been assigned to a group. The
algorithm has been modified so that at each stage the machine loading can be
evaluated and a check made that no proposed grouping violates any restriction
on machine availability. Table 3 shows part of a machine loading print-out.
By inspecting the component grouping and machine load print-out obtained
at successive levels of similarity coefficients in the cluster analysis process, the
groupings providing the best practical component families may be selected.
To reduce the volume of print-out to be studied the programme may be re-
quested to provide the print-outs at intervals, of say 0,05, in similarity
coefficient, not every time a component is added to a family, or every time two
small families are combined. In this case selecting the optimum families may
involve some minor adjustment to the families in thc print-out which usually
can be deduced by inspection.
408 A. 8. Carrie

Table 3, Machine load printout

GRO\Jg MEt,qERS TOT~l IHCHINE WORK LOADS MtNUTH


NUMBER SVtlBoL CODE SET-Up RU'.

,,~
80 241 , 62 Ho. 10~,
88 99/1 146 Ho. Hl,
2
260
60 360 196 720. 60U.
62 6'" 149 180. 12,
3
201 181 30 183 190
3S 191 115 193 211
17' 151 141 \44 13 1
Downloaded by [Universitaets und Landesbibliothek] at 06:44 05 September 2013

135 13 4 131 ,, 9 i'1 H


116 212 213 i!1 J a,Y
,4 24 1 238 321 J.t 7
20 21 22 2J :163
31 0313 16 24420. 6620' ,
38 03/1 14 15000. 205 oy •
39 03/4 19 I O~O. 11]4,
43 03/" 80 5040. 116U,
50 A6/ 8 91 ,0080. 9116u.
51 061 S ,66 4680. 142ll,
4
3" 70 2640. 7'82,

5
"
87
03/S
25/' 152 Ho. HOs,

141 140 136 346 1')J


,,7 H2 4 13J 50
4° 142 202 ,91 196
2411 \70 159 251 ,,4
254 192
17 03/3 76 ,,520. 231174,
33 03/1 74 8llo. 4173 8,
39 03/4 19 72,1. 2HU,
40 03/ 9 84 2'60. I 26U,
43
48
50
OJ/1
06/2
A6/ s
,,,
80
97
3240.
2880.
7200,
9000,
420U,
8n~,
51 06/' 166 396,). 819 U,
1.6 24/1 58 o. 5 S,
2400. 18, 5Y ,
79
81
2414
2'4/ 6 "
63 12,1. 302 4,

Application to planning functional layouts


In a functional layout all machines of each function are placed together.
Mutetia.l flow may be taken into account by placing close to the machines of
each function the machines of these functions to or from which a large amount
of material flows. The travel chart, showing the amounts of material moving
from each function to cach of the other functions, can be used to facilitate
preparing the layout plan. Intuitively it is clear that a good procedure is
scanning the travel chart to find the largest entry, and diagrammatically
placing the two functions concerned adjacent to each other, and then repeatedly
rescanning the chart for the next largest entry and adding the appropriate
function, until all functions have been added. The diagram so formed must
then be adjusted to give each function its required area and to give the arrange-
ment a realistic overall shape.
The application of this simple principle can become a complex manual task
in a problcm involving more than about 20 functions since the relationships
between each and every other must bc considered. Before the assistance of a
Numerical taxonomy applied to group technology and plant lnyout 409

computer can be obtained the procedure must be defined in a computer pro-


gramme. This is no easy task. Of existing programmes, such as CRAFT (5)
and CORELAP (6), none is wholly satisfactory or readily available outside a
few research or consulting organizations. Both CRAFT and CORELAP treat
the floor area of the plant as made up of a large number of rectangular blocks
and assign to each block one of the functions to be included. The number of
blocks of area required to represent each function is proportional to its floor
space requirement. Both programmes ensure that all the blocks representing
a function are adjacent to one another. Both programmes have the disadvan-
tages that (a) the shape of the outline of the space assigned to each function
may be highly irregular and (b) the problem of arranging the machines within
Downloaded by [Universitaets und Landesbibliothek] at 06:44 05 September 2013

each functional area remains unsolved. CORELAP may also produce a highly
irregular building outline shape. As Muther and McPherson (1970) have pointed
out a purely analytical technique that can be programmed on a computer will
never, on its own, fully solve a plant layout problem. There are always less
quantifiable factors to be taken into account which call for the subjective
judgement of the industrial engineer. The computer programme should not
attempt to work out a solution to a degree of detail which restricts or ignores the
industrial engineer's role. Instead it should present results in a way which
encourages and simplifies his role, and it should do so by a logical process which
he understands and in which he has confidence. The techniques of numerical
taxonomy provide an analytical procedure remarkably similar to the intuitive
one proposed above.
In describing the principles of cluster analysis, it was mentioned that objects
to be classified which were highly similar to one another could be thought of as
being close together in a multi-dimensional space. The classification process
formed clusters of near-neighbours in that space. Applying this to plant
layout, the objects to be classified are the machines, functions or departments
involved in the problem. It is necessary to define their characteristics in such
a way that those functions which should be placed close together in the plant
layout become near neighbours in the multi-dimensional space representing
the cluster analysis stage of the classification process. Since functions should
be placed close together, if a large amount of work flows between them this can be
achieved by defining the similarity coefficient between pairs of functions in
terms of the amount of work flowing between them. Therefore the travel
chart can be used as the data matrix from which these similarity coefficients
are calculated. The results of the classification process can be presented in the
form of a minimal spanning tree. The cluster analysis procedure in identifying
groups of near neighbours in multi-dimensional space, may be requested to
produce a list of the nearest neighbours of each object. The minimal spanning
tree is constructed from this list, and the branches of the tree join each function
to the one that should be placed closest to it in order to minimize material
handling effort. The tree may be used directly as a schematic plant layout
diagram, requiring only the adjustments of floor area and overall shape men-
tioned above.
This approach was used in a layout study for an earthmoving equipment
manufacturer. The company proposed to transfer the manufacture of a trans-
mission unit from the United States to one of their Scottish factories. The
layout study was concerned with the factory extension which would have been
410 A. 8. Oarrie

Fig. 6
Downloaded by [Universitaets und Landesbibliothek] at 06:44 05 September 2013

Schematic plant layout diagram-earthmoving equipment manufacturing study.


Numerical taxonomy applied to group technology and plant layout 411

necessary to accommodate the required additional machinery. As a first stage


of analysis a search for families of components for group production was carried
out following the methods described above. The resulting families were con-
sidered to be insufficiently well defined for group production to be effective,
and it was decided to adopt a functional type layout. All the components were
treated as one group and the material flow between machines was examined.
Whereas a functional layout normally treats all machines of a process, e.g.
turning or milling, as a group and places them together in the layout (although
a finer division to the level of centre lathes, turret lathes, horizontal mills,
vertical mills is often used) it was considered that a better flow pattern would
be obtained by distinguishing between one size or specification of machine of
Downloaded by [Universitaets und Landesbibliothek] at 06:44 05 September 2013

similar type and another. To provide the required capacity more than one
machine of some specifications would be required, but these would all be placed
together in the layout.
The minimal spanning tree for material handling between machines was
derived, and adjusted to give the schematic layout diagram as shown in fig. 6.
Each circle represents a machine, or group of machines of the same specification,
with an abbreviation of the machine's name and code number, and the size of
the circle is in proportion to the floor area required. The main paths of flow
forming the minimal spanning tree are also shown.
INSP2 near the centre of the figure is the department where completed
components and sub-assemblies are inspected or where partially machined
components are inspected prior to heat treatment. Situated close to this
department, is the assembly department which is also the area where com-
ponents are stored, if necessary, prior to assembly. The completed unit moves
up to the right through testing, washing and packing to the despatch area. All
the heat treatment processes are together at the central right-hand side. These
are close to INSP2, through which partly machined components requiring heat
treatment must pass. After a component has been treated it must be inspected
at INSP3 before returning for further machining. This further machining is
usually drilling or grinding so that, some machines of these types are located
close to INSP3 at the lower right. On the other hand since grinding is usually
a final operation most of the grinding machines are situated at the upper left,
near the DAWSON automatic washing machine, which itself is adjacent to final
inspection. Manual deburring at a BENCH, also usually a final operation, is
also situated near the washing machine. Forgings and castings are checked,
before processing, at INSPI, to the right of the diagram. They then move into
the heat treatment section or to the machine shop for milling and drilling.
Several splined shafts are components of the transmission unit and a cell for
these is at the lower left, consisting of SAW, face and centering machine, (F&C),
copy lathe (P5) and HOB. There are also several gears and the nucleus of a cell
for these is in the centre of the diagram consisting of a hob, (HOBPH), deburring
machine (NTG) and a gear shaving machine (SHAVER). The Wickman
automatic lathe 'VICKAUTO is situated near INSP2 since many of the
components produced by it are completed in a single operation. The lathes not
mentioned already are in a functional group at the bottom centre of thc diagram.
This layout is an interesting compromise between the group and functional
types of layont. By examining machine types in detail it affords a better
overall flow pattern than the normal functional layout.
412 A. S. Carrie

Some theoretical comments on this approach to functional-type layout


A travel chart showing material movement between N machines will have
N x (N -1) entries if the entries on the main diagonal are ignored. These
cntries may be taken in pairs to show the total movements between machines
irrespective of direction of flow, yielding iN x (N - I) relationships. The
minimal spanning tree (l\iS'I') selects the (N - 1) most significant of these
relationships, thus including 2 x (N -1) of the travel chart entries. The
significance of an entry is not merely dependent on the magnitude of the entry,
but on its magnitude in relation to the other entries for the same machine. In·
this way the MST links each machine to the one with which it has the most
. material flow. The MST approach may be criticized on the grounds that
Downloaded by [Universitaets und Landesbibliothek] at 06:44 05 September 2013

although it shows the most important paths of work flow (for example in fig. 6
2ACCH is linked with 7DWARD, VlOB, BROACH and INSP3) it does not
show any paths of work flow with other machines, but only describes them as
, not of most significance'. Some tests were carried out to assess the serious-
ness of this limitation.
Thc first and most significant point is that many of the N x (N - I) entries
arc zeroes. In fact in the cases analysed the average proportion of zero entries
was 78%. Furthermore, of the 22% of entries that were non-zero it was found
that more than half contained small trivial values. The important entries
represented only 10% of all entries. With such a small number of important
entries the MST can select a very high proportion of the significant entries.
Table 4 shows the proportions of material handling effort accounted for by the
l\fST in the cases tested, (average 65%) and shows that the success of the 1\'18T
increases with the number of zero entries with the number of important
relationships. It also suggests that the proportion of zero entries increases
with the number of machines, rendering the approach relatively more effective
in relatively complex problems. The problem of fig. 6 is number one in
table 4.

Table 4. Effectiveness of minimal spanning tree in functional layout design

Travel chart examined


I 2 3 4 5
Number of machines 59 27 20 17 10
---
% of entries which were Important I 5 11 8 24
Trivial I 6 Jl 14 25
Zero 92 89 78 78 51
% of material handling effort Highly
accounted for in M8T of important 100 100 100 94 67
entries which were Important 98 83 86 88 64
Non-zero 59 70 63 75 59

Note: Entries which accounted for more than I % of material handling were
considered important, those of more than 2% highly important and those of less than
I % were considered trivial.
Numerical taxonomy applied to group technology and plant layout 41 3

It is not suggested that the diagram derived from the MST represents the
final stage or last word in designing a functional layout. It is suggested
however, that this approach takes the purely analytical process to an appro-
priate stage before the industrial engineer brings his subjective judgement to
bear on the less quantative aspects of the problem, and that it presents informa-
tion to him in a suitable form.

Conclusions
This paper has sought to describe the principles of numerical taxonomy and
to show that it is a technique deserving of the attention of industrial engineers.
Downloaded by [Universitaets und Landesbibliothek] at 06:44 05 September 2013

Its relevance to group technology and plant layout may be summarized as


follows:
(I) It provides a means of assessing whether group production methods will
be effective, or whether a functional type of layout should be adopted.
(2) It provides a means of forming component families for group production
methods, based on the principles of production flow analysis.
(3) It provides a means of analysing work flow for the development of a
functional type layout.
(4) It does not complicate the task of reconciling analytical procedures with
subjective judgement, but can present its results in various forms so
that the industrial engineer may select the one which is most useful.
The techniques for defining similarity between items and for cluster analysis
described in this paper are only few among many available in numerical
taxonomy. They were selected on account of their simplicity, availability, and
the appropriateness of their methods of presenting results. The range of
techniques available can be learned from a study of the proceedings of the recent
conference on numerical taxonomy at which Wishart's paper (1969) appeared.
'York on the development and application of the techniques to production system
design problems is continuing.

ACKNOWLEDGMENTS

The author wishes to express his appreciation of the assistance of the many
firms who have cooperated on these studies, of the students who have worked
on them under the author's direction, and of Professor D. S. Ross, Head of the
Production Engineering Departmcnt at the University of Strathclyde.

Appendix
Sub-routine GRPANALYS1S
Listing-see fig. 7.
Input parameters
(I) Data information:
13 the disc file channel number.
Nl the number of components for which information is, or could be,
held.
414 A. S. Carrie

:Fig,7

SUBROUTINE ~RPANALYSIS(13.N1.N2.N3.NE.NMX.NMX1.NP.NM.K1.NTAB.NA2.
1 MA.MA2,L1.L2,L3.L4.SC,IM. I1.12.14.MA3)
DIMENSION NTAB(N1),NA2(NP),MA(N2).MA2(NMX).L1(NMX).L3(NMX1).
1 L2(NMX),L4(NMX1).SC(NMX).MA3(NM)
C CALCULATE ADDRESS OF FIRST WORD OF DATA MATRIX ON DISC STORE
K2=K1+N3+N3'NE'N1
c CALCULATE ADDRESS OF FIRST WORD OF SIMILARITY COEFFICIENT MATRIX
K3=K2+NM'NP
C CALCULATE ADDRESS OF FIRST WORD OF DENDOqRAM ON DISC STORE
K4=K3+NMX'NMXI2
c CONSTRUCT COMPONENT - MACH INE DATAMATRI.X .ON DI S,C
CALL BETCOM(13,N1.N2.N3.NE.K1.K2,NP.NM,NTAB.L1.NA2.MA.L3.L2)
c
Downloaded by [Universitaets und Landesbibliothek] at 06:44 05 September 2013

PRINT INITIAL COMPONENT - MACHINE MATRIX


CALL MATPRINT(13,K2.NP.NM,L2,NA2)
C IF REQUESTED CONSTRUCT AND PRINT INITIAL MACHINE - COMPONENT MATR!
IF(I~,) 1,1.3
1 CALL BETMAC(13.N1.N2,N3.K1.K3.NE.NP,NM.NTAB.L4.NA2.MA.L3,L2)
CALL MATPRINT(13.K3.NM,NP.L2.MA3)
c CALCULATE SIMILARITY COEFFICIENT BETWEEN EACH PAIR OF COMPONENTS
c AND STORE ON DISC
3 CALL SCOEF1(11.13.K2.K3.~PSNM,L1.L2.L4.SC.12)
C COMPONENT qROUPINq BY ROSS S CLUSTER ANALYSIS AL40RITHMS
CALL PRIMTREE(13,K3,NP,-1.D.L1.SC.L2)
CALL SLCA(13.K4,NP.NP+1.-1.0,O.05.L1.SC.MA2.L2,L3.L4)
C RECONSTRUCT 'COMPONENT LIST IN NEW ORDER
DO 12 1=1.NP
12 MA2( I )=NA2(MA2( I»
DO 13 1=1, NP
13 NA2(1)=MA2(1)
C CONSTRUCT AND PRINT INTERMEDIATE COMPONENT - MACHINE MATRIX
CALL BETCOM(13.N1,N2.N3.NE.K1.K2,NP.NM.~TAB.L1.NA2.MA.L3.L2)
CALL MATPRINT(13.K2.NP.NM.L2,NA2)
c CONSTRUCT (INTERMEDIATE) MACHINE - COMPONENT DATA MATRIX
CALL BETMAC(13.N1.N2,N3.K1.K2.NE.NP.NM.NlAB.L4.NA2,MA.L3.L2)
c IF REQUESTED PRINT INTERMEDIATE MACHINE - COMPONENT MATRIX
IF(IM) 5.5.7
5 CALL MATPRINT(13.K2,NM.NP.L2.MA3) .
C CALCULATE AND STORE SIMILARITY COEFFICIENT MATRIX BETWEEN MACHINES
7 CALL SCOEF1(11.13.K2,K3.NM.NP.L4.L1.L2,SC.14)
C MACHINE qROUPINq BY ROSS'S CLUSTER ANALYSIS ALqORITHMS
CALL PRIMTREE(13,K3,NM.-1.0.L1.SC.L2 )
CALL SLCA(13,K4.NM.NM+1.-1.0,O.05,L1.SC.MA2.L~.L3.L4)
c RECONSTRUCT MACHINE ADDRESS ARRAY FOR NEW· ORDER
DO 6 1=1,N2
IF(MA( I)) 6.6,B
B'DO 9 J=1,NM
IF(MA2(J)-MA( I» 9,2,9
9 CONTINUE
2 MA(I)=J
6 CONTINUE
C IF REQUESTED CONSTRUCT AND PRINT FINAL COMPONENT - MACHINE MATRIX
IF( 1M) 10.10,11
10 CALL BETCOM(13.N1.N2.N3.NE,K1,K2,NP,NM.NTAB.L1.NA2.MA.L3.L2)
'CALL MATPRINT(13.K2.NP,NM,L2.NA2)
C CONSTRUCT FINAL MACHINE - COMPONENT MATRIX
11 CALL BETMAC( 13,N1.N2.N3,K1.K2.NE,NP.NM.NTAB,L4.NA2.MA.L3.LZ)
C RECONSTRUCT MACHINE ROW IDENTIFICATION NUMBERS IN NEW ORDER
DO 15 1=1,NM
15 L1(I)=MA3(MA2(1»
C PRINT FINAL MACHINE - COMPONENT MATRIX
CALL MATPRINT(13.K2.NM.NP.L2.L1)
RETURN
END
Listing of Subroutine GRPANALYSIS.
Numerical taxonomy applied to group technology and plant layout 415

N2 the number of machines or processes which may be required for


manufacture of the components.
N3 the number of words in each record for each component.
NE the number of records for each component.
K1 the address on disc file for the first work of the data file.
NTAB the number of operations on each component.

(2) Data on the analysis to be performed:


NP the number of components to be analysed.
NM the number of machine or process categories III the machine-
component matrix.
Downloaded by [Universitaets und Landesbibliothek] at 06:44 05 September 2013

NMX the maximum of NP and NM.


NMX1 one more than NMX.
NA2 the list of components to be analysed in the order in which they
are to appear initially in the matrix.
MA the row number of the machine-component matrix to which
each machine or process is to be assigned.
MA:~ the identification number to be applied to each row of the
matrix (i.e. process identification).

(:~) Control variables:


II to select the method of defining the similarity coefficients.
12 to select the form of output, if any, for the component similarity
matrix.
14 to select the form of output, if any, for the machine similarity
matrix.
1M to select which machine-component matrices are to be printed.

(4) Working area arrays: MA2, LI, L2, L3, L4, SC.

On exit
NA2 and MA specify the revised order of components and machine cate-
gories, and L 1 gives the revised order of process category identification numbers.

Cot article conatnte qu'une technique est necesaa.ire, qui puisse evu.luer si un syeteme do
production fonctionnel ou it cellule de groupe convient. Io mieux it un cas spectfique, ct qui puisse
6galemcnt developper Ie systeme approprtc. L'article (Merit In technique de taxonomic nurneriquo
et mont.re comment OIl pout l'appliquer it In technologic de groupe auaai bien qu'a In planificntion
do l'installation.

On presente un programme d'ordlnnteur pour analyser Ie cheminemcnt de In production.


On mantre ensuitc que In methode presentee par Burbidge, qui nnulyse lo cheminernenf de ILL
production par Ia mate-ice composontc-machine, n'LL qu'uno valeur Jiruiteo dans les cas pratiques
et on propose WlO methode amelioree.

On deer-it, enfin l'ut.iliaation de In taxonomie numer-iquc pour devolopper un schema d'inetella-


tion de typo fonctionnel.
4lO Niunericai laxonomy applied to group technology and plant layoul

Diose Abhundf ung weist des Bedurfnis nach einem Verfahren nnch, auf Grund desscn man
voranschlagcn knnn- ob fur- einen best.immten Fall ein funktionelles oder ein Gruppenzell-Produk-
cioneayetcm voranzlohcn ist und daruberhinaua des betroffene System wei tel' zu entwickeln lat.
Dol' Bericht beachreibt dna Verfahren del' numerischen Taxonomic und wie dieses sowohl fur die
Gmppcntechnologie ale auch fur die Auslcgung einer Anlage angewendet werden kann. Es wird
cin Oomputerprogrcnun fur die Produkt.ions-Iluflauawert.ung vorgelegt. \Veitcrhin wird erwiesen,
dnl3 die von Burbidge vorgelogto Einheitenmaschinen-l\latrizenmethode fur die Auewertung des
I'rodukt.ionafluesea in praktischer Hinsicht einen begrenzten Wert hat, und ea wird cine verbesserte
Methode vorgcschlagcn. Der Einsatz der numerischen Taxonomic in der Entwicklung einer
funktionsgcartetcn Auslegung wird ebenfalla beschrieben.

REFERENCES
BUFFA, K S" ARMOUR, G, C" and VOLLMAN, '1', K, 1964, Allocating facilities with
Downloaded by [Universitaets und Landesbibliothek] at 06:44 05 September 2013

CRAFT, Harvard Business Review (March-April),


BURBIDGE, J, L, 1969, An introduction of group technology, Proceedings of Seminar
on Group Technology, Turin; 1971, Production flow analysis, Proceedings of
Seminar on Group Technology, International .ll1anagement Centre, Turin 1969
(Editor J, L, Burbidge) also The Production Engineer, 50, (4/5), April/May,
CAUlUE, A, S" 1971 a, Correlation theory applied to work flow measurement, Int. J,
Prod, Res" 9, 209; 1971 b, The design of production systems, Proceedings of
Second International Conference on Product Development and ll1anufacturing
Technology, University of Strathclyde, April,
EDWARDS, G. A. R, 1972, Correspondence to the Editor, Prod, Enqr, 51, (7/8),
HOLLIER, R H" 1963, The layout of multi-product lines, Lnt. J. Prod, Res" 2,47,
HOI"I"IER, R, H" and CORLETT, K N., 1966, Work flow in batch production, Problems
of Progress in Industry, No, 18, H,lVLS,O.
HOLSTEIN, \\T. K" and BERRY, \V. L" 1970, Work flow structure: an analysis for
planning and control. 11fgmt Sci., 16, 324,
LE~], R, C" and MOORE, J. M., 1967, Corelap-computcrised relationship layout
planning, J. uul. Enqnq, 18, (3), March.
MACAULEY, J" 1970, A study of machine grouping, Research project report for
Diploma in Operational Research, University of Strathclyde, 1970; 1972,
Machine grouping for efficient production, Prod, Engr, 51, (2),
!\IUTIIER, R" and MCPHERSON, K" 1970, Four approaches to computerised layout
planning, Ind, Engng, February,
Ross, G, J. S" 1969, Algorithms AS13, ASI4 and AS15, Appl. Statist" 18, (I).
SOKAL, R. R, and SNEATH, P, H. A" 1963, Principles of Numerical Taxonomy
(Freeman & Co.).
WISHART, D" 1968, Mode analysis, Proceedings of Colloquium in Numerical Taxonomy
at St Andrews, Scotland, September, edited by A, J, Cole (Academic Press),
1969,

Das könnte Ihnen auch gefallen