Академический Документы
Профессиональный Документы
Культура Документы
k=1
r
ik
= 1, v
-
i = 1, 2, n, (7)
where n i s the overall number of resource
elements. Equati on (7) shows that each for m
generati ng scheme f
i
i n F belongs to only one
RE and that there i s no overlappi ng between
the RE; hence i f the deci si on-maki ng process
i s based on REs then there are no alter na-
tives to be consi dered at thi s level of for mal-
i zati on of the capabi li ti es of a machi ni ng
faci li ty[13].
The break-up of the machi ni ng faci li ty i nto
resource elements i s summari zed by the
followi ng algori thm:
1 Step 1. Dene the capabi li ty of each
machi ne tool usi ng FGS.
2 Step 2. Select FGS f
i
.
3 Step 3. Select f
j
(j i).
4 Step 4. I f m
ik
= m
jk
for each machi ne M
k
(k = 1, 2, , m) then cluster together f
i
and
f
j
.
5 Step 5. Repeat steps 2,3 and 4 for all i, j = 1,
2, , n.
6 Step 6. Dene each cluster of FGS as a RE.
7 Step 7. Usi ng the set of REs, represent each
machi ne by i ts REs.
Component dat a model
The descri pti on of the processi ng capabi li ti es
of a machi ni ng faci li ty i n ter ms of resource
elements provi des a basi s for faci li ty speci c
deni ti on of the component processi ng
requi rements. Components can be repre-
sented by uni que sets of resource elements
li nki ng them to sets of possi ble machi ne
alter natives wi thout pre-selecti ng any of the
machi ne tools.
The component set i s for mally dened i n
the Eucli dean space C. Each component i s
descri bed as an element i n C wi th co-ordi -
nates representi ng the processi ng requi re-
ments i n ter ms of resource elements. Compo-
nents are represented by vectors
C
k
= c
1
k, c
2k
, , c
nk
(8)
where c
ik
(i = 1, 2, , n; k = 1, 2, , c) i ndi cates
the relati onshi p between component C
k
and
resource element R
i
as:
c
ik
={
1, i f component C
k
requi res RE R
i
,
0, i f component C
k
does not
requi re RE R
i
(9)
Fi gure 2
Resource element denition
[ 218 ]
Nabil N.Z. Gindy and
Tsvetan M. Ratchev
Cellular decomposition of
manufacturing facilities using
resource elements
Integrated Manufacturing
Systems
8/ 4 [1997] 215222
Cel l format i on met hodol ogy
As a result of the break-up of the capabi li ty
of the machi ni ng faci li ty i nto resource
elements, the groupi ng task i s transfor med
i nto opti mum parti ti oni ng of the
component set and the machi ni ng faci li ty
by usi ng resource elements to represent
both the component processi ng requi re-
ments and the capabi li ti es of the machi ni ng
faci li ty.
The parti ti oni ng of the component set and
the machi ni ng faci li ty are carri ed out con-
currently usi ng an i terative clusteri ng
approach. At the fi rst i terati on the compo-
nents are clustered to produce capabi li ty
patter ns whi ch defi ne the boundari es of the
machi ni ng cells i n ter ms of resource ele-
ments. At the second i terati on the resource
elements belongi ng to each cell are trans-
lated i nto physi cal machi ne tools (see Fi gure
3). The procedure conti nues unti l an opti -
mum parti ti on of the machi ni ng
faci li ty/ component set i s reached.
Component cl ust eri ng procedure
The cell for mati on approach i s based on the
divi sive fuzzy C-means (FCM) clusteri ng
algori thm[14]. The FCM algori thm i ncludes
the followi ng three i terative clusteri ng
steps:
1 I ni ti ali ze the membershi p functi on
ij
, of
component C
i
to group G
j
such that:
c
i=1
ij
= 1 (10)
2 Compute the fuzzy groupi ng centroi ds G
i
for i = 1,2, , gdened as wei ghted sums of
all data poi nts i n the set
where n i s the total number of components
and f > 1 i s a fuzzi ness i ndex[14].
3 Update the fuzzy membershi ps
ij
usi ng:
where e
2
(C
i
, G
j
) i s the Eucli dean metri c
nor m[14].
The three steps are repeated unti l a stable
component/ machi ne parti ti on i s achi eved for
each parti cular number of component
groups/ machi ni ng cells.
Val i dat i on of component cl ust ers
The best faci li ty parti ti on i s selected usi ng
a cluster vali dati on measure. One of the
i mportant i ndi cators for effi ci ent cellular
manufacturi ng i s the break-up of the machi n-
i ng faci li ty i nto product based cells of di s-
si mi lar producti on equi pment[15]. Therefore,
the vali di ty measure used for parti ti oni ng the
component set i nto groups and the machi ni ng
faci li ty i nto cells needs to be dened i n such a
way as to ensure the for mati on of component
groups wi th maxi mum si mi lari ty between
the component processi ng requi rements and
machi ni ng cells wi th maxi mum diversi ty, i .e.
mi ni mum overlappi ng i n ter ms of the
repeated machi nes between cells.
I n the context of the cell deter mi nati on
task, the vali dati on cri teri a can be for mali zed
as (see Fi gure 4):
maxi mum compactness of the machi ni ng
cells i n ter ms of capabi li ty;
maxi mum compactness of the component
groups i n ter ms of processi ng
requi rements;
mi ni mum overlappi ng of the component
groups i n ter ms of processi ng
requi rements;
mi ni mum overlappi ng of the machi ni ng
cells i n ter ms of capabi li ty.
A cluster vali di ty measure proposed i n [10] i s
closely related to the above cell for mati on
cri teri a, owi ng to i ts capabi li ty to assess the
cluster tendency i n ter ms of both compact-
ness of component groups and di fferenti ati on
between the clusteri ng centroi ds. Therefore,
i t can be used as a vali di ty measure to i ndi -
cate the opti mum parti ti on of the machi n-
i ng faci li ty.
The vali di ty measure i s dened as rati o of
the overall compactness of the parti ti on
to the separati on between the cluster s:
s=
I I
s
The overall compactness of the parti ti on i s
dened as:
=
1
g
q
j=1
j
(13)
where
i
= (
i
/ n
i
) i s the average devi ati on of
component group G
i
andgi s the total number
of groups[10].
The cluster separati on measure s, i ndi cat-
i ng the level of overlappi ng between the clus-
ters, i s dened as the mi ni mum di stance
between any two clusters of the parti ti on:
s = mi n
ij
(b
ij
)
2
(14)
where b
ij
= e(G
i
,G
j
) i s the Eucli dean di stance
between group centroi ds G
i
and G
j
.
The vali dati on measure i s used for selec-
ti on of the opti mum faci li ty parti ti on.
Mi ni mum vali di ty measure S i ndi cates
ij
j i
f
j i
f
i
c
e C G
e c G
=
1
1
12
2
1
1
2
1
1
1
( , )
( , )
( )
( )
( )
G
C
i
ij
f
j
j
n
ij
f
j
n
=
=
=
( )
( )
( )
1
1
11
[ 219 ]
Nabil N.Z. Gindy and
Tsvetan M. Ratchev
Cellular decomposition of
manufacturing facilities using
resource elements
Integrated Manufacturing
Systems
8/ 4 [1997] 215222
opti mum parti ti oni ng of components
i nto groups and the machi ni ng faci li ty i nto
cells. I t i s also used for measuri ng the
stabi li ty of the parti ti ons at each i terative
step i n the FCM algori thm, whi ch si gni -
cantly si mpli es the calculati on
procedure[10,14].
The vali di ty measure i s dened at two lev-
els usi ng resource elements (Sr) and usi ng
machi nes (S
m
). At the i terati on for each num-
ber of component groups/ machi ni ng cells a
Sr i s used to select stable parti ti on before
machi ne selecti on. Once a stable resource
element based parti ti on i s achi eved and the
Fi gure 3
Facility decomposition methodology
[ 220 ]
resource elements are transfor med i nto
machi ne tools, the nal secti on of a stable
faci li ty parti ti on i s based on vali di ty measure
S
m
.
Machi ne sel ect i on
At each i terati on the machi ne tools for each
cell are selected by matchi ng thei r capabi li -
ti es to the cell boundari es descri bed i n ter ms
of resource elements. The machi ne selecti on
i s based on a wi de vari ety of cri teri a and may
i nclude requi rements such as: mi ni mum
vari ety of machi nes i n the cells; mi ni mum
transport movements i n the cell; maxi mum
uti li zati on; opti mum level of concentrati on of
the operati ons (e.g. selecti on of machi ni ng
centres agai nst dedi cated machi nes), etc.
A fundamental advantage of usi ng REs i n
cell for mati on i s the abi li ty to translate the
cell capabi li ty boundari es i nto di fferent sets
of machi ne tools. Thi s allows multi ple
choi ces i n cell congurati on and selecti on of
the most effi ci ent combi nati on of worksta-
ti ons i n ter ms of machi ne capabi li ty and
capaci ty uti li zati on. Research i s ongoi ng i n
developi ng a deci si on-maki ng module to
support the allocati on of machi ne tools to the
RE based cell boundari es under a vari ety of
capaci ty constrai nts. The results so far sug-
gest that by taki ng the deci si ons at sub-
machi ne level usi ng resource elements a
more accurate esti mate of the actual capaci ty
requi rements of parti cular machi ni ng capa-
bi li ti es can be provi ded and, therefore, a more
effi ci ent faci li ty decomposi ti on can be
achi eved.
I n thi s paper an i llustrative example i s
presented where the selecti on strategy
emphasi zes the mi ni mi zati on of the vari ety of
machi nes and transport movements i n the
cells. To sati sfy that requi rement, machi ne
tools are selected on the basi s of maxi mum
capabi li ty, i .e. pri ori ty i s given to machi nes
wi th the maxi mum number of resource ele-
ments, i ncluded i n the cell boundari es.
Each machi ne M
k
M i s, therefore, evalu-
ated i n ter ms of number of REs common to
the group centroi ds (representi ng the cell
capabi li ty boundari es). Machi nes are then
clustered around the group centroi ds, substi -
tuti ng correspondent REs. The process i s
based on substi tuti ng the maxi mum number
of REs at each step by i ncludi ng the most
capable machi ne tool. The selecti on proce-
dure i s completed once all REs i n the compo-
nent parti ti on are substi tuted by machi ne
tools.
Cel l val i dat i on
The i terative cell for mati on procedure i s
repeated starti ng wi th two component
groups/ machi ni ng cells i ncreasi ng thei r
number unti l a stop value i s reached
beyond whi ch the decomposi ti on i s consi d-
ered i nfeasi ble[10]. At each step machi ne tool
based vali di ty measure S
m
i s calculated i n
order to dene a stable parti ti on
g
(where g
i s the current number of groups/ cells).
A faci li ty parti ti on wi th mi ni mum vali di ty
measure S
m
i s selected as opti mum:
where gi s a stop-value of the cell number.
The vali di ty measure S
m
has a monotoni c
decreasi ng tendency as the number of cells
reaches a value close to the number of compo-
nents. Thi s, however, i s not a seri ous prob-
lem, si nce i n practi ce, the feasi ble number of
clusters i s much smaller than the number of
components[10]. Heuri sti c methods can be
used to deter mi ne a stop-value beyond
whi ch groupi ng i s consi dered unfeasi ble. One
of the possi ble approaches i s to dene the
value for S
m
for g= 2,3,,n 1 and select the
starti ng poi nt of monotoni c decreasi ng ten-
dency as the maxi mum number of groups to
be consi dered.
Faci l i t y decomposi t i on al gori t hm
The faci li ty decomposi ti on algori thm for cell
deter mi nati on usi ng resource elements can
be summari zed as:
1 Select i ni ti al groupi ng centres.
2 Calculate the component membershi p
functi on.
3 Calculate the new groupi ng centres and
RE membershi p functi on.
4 Calculate RE based vali di ty measure Sr.
mi n mi n
( )
2
15
g g
S
m
g
Fi gure 4
Cell validation criteria
Nabil N.Z. Gindy and
Tsvetan M. Ratchev
Cellular decomposition of
manufacturing facilities using
resource elements
Integrated Manufacturing
Systems
8/ 4 [1997] 215222
[ 221 ]
Nabil N.Z. Gindy and
Tsvetan M. Ratchev
Cellular decomposition of
manufacturing facilities using
resource elements
Integrated Manufacturing
Systems
8/ 4 [1997] 215222
5 Repeat steps 2-4 unti l the stable parti ti on i s
achi eved (Sr = mi n).
6 Select machi ne tools for each cell whi ch
match the requi red resource elements by
the component groups.
7 Calculate machi ne based vali di ty measure
S
m
.
8 I ncrease the number of groups g = g+1.
9 Repeat steps 1-8 unti l g = g.
10 Select the parti ti on wi th mi ni mum S
m
.
Experi ment al resul t s
The decomposi ti on methodology was experi -
mentally tested usi ng data from a number of
i ndustri al cases. The case i llustrated here i s
based on set of 50 components and machi n-
i ng faci li ty of 45 machi ne tools. The compo-
nent set i s a representative sample from the
database of a large i ndustri al company. Com-
ponents are selected i n the same si ze range
and wi th small producti on volumes so that
thei r for m generati ng requi rements are
consi dered to be the major clusteri ng cri te-
ri on for faci li ty decomposi ti on. I n order to
vali date the approach the machi ni ng faci li ty
i s decomposed twi ce usi ng component-
resource element and component-machi ne
tools representati on of the component route-
i ngs.
The results of the decomposi ti on are
summari zed i n Fi gure 5. The opti mum parti -
ti on i s deci ded by the machi ne tools based
compactness and repeti ti on measure S
m
calculated for each parti ti on. The best parti -
ti on i s found to be the one wi th five groups
and S
m
= 0.022. The appearance of some of
the groupi ng pri mi tives i n more than one
centroi d, whi ch nor mally would be consi d-
ered as bottle-neck elements, i n thi s case
could be i gnored owi ng to the fact that all
repeated REs can be translated i nto di fferent
machi nes; hence, at machi ne level there i s
no repeti ti on of machi ni ng resources (i n
some cases li mi ted repeti ti on can be
unavoi dable because of the shortage of
machi ni ng resources).
The results of the faci li ty decomposi ti on
usi ng machi nes whi ch were i ni ti ally
selected i n the component process plans are
shown i n Fi gure 6. I t can be observed that
the best parti ti on i s agai n the one wi th five
groups (S
m
= 0.029) but because the
machi nes have been selected before the
groupi ng they do not reflect the commonal-
i ty between the components i n the groups;
therefore component devi ati on from the
centroi ds i s hi gher, whi ch when translated
i nto cell perfor mance means lower equi p-
ment uti li zati on and hi gher transport ti mes.
There i s also a constant trend of repeated
machi nes (i .e. machi nes requi red by more
than one cell), whi ch at thi s level of for mal-
i zati on of component requi rements cannot
be avoi ded.
I n the case of RE based cell deter mi nati on
the overall number of machi nes selected for
the five cells i s consi derably smaller (see
Fi gure 7) than the number of machi nes
selected when component-machi ne for mal-
i zati on i s used. The di fference i s parti ally
based on the subjective selecti on of compo-
nent routi nes by the planners i n the second
case. However, the fact that machi nes are
selected for groups of components rather
than on a component-by-component base
shows that faci li ty decomposi ti on based on
resource elements provi des a wi der choi ce of
selecti on alter natives and leads to i mproved
machi ne selecti on.
As the compari son of the graphs shows
(see Fi gures 5, 6 and 7), the cells defi ned
usi ng REs have better compactness and repe-
ti ti on values i n ter ms of machi nes, are
defi ned usi ng lesser number of machi ne
tools and there i s no machi ne repeti ti on. The
groupi ng based on pre-selected machi nes
has hi gher vali dati on measure i ndi cati ng
hi gher devi ati on of the components i n the
groups (whi ch i n most cases translates i nto
lower machi ne uti li zati on) and leads to
i ncreased number of repeated machi nes
(bottle-neck elements).
Concl usi ons
The resource element concept proposed i n
thi s paper provi des a process capabi li ty
descri pti on whi ch i s both machi ne i ndepen-
dent and faci li ty speci c. The resource ele-
ments are used to dene component process-
i ng requi rements uni quely wi thout rejecti ng
any possi ble machi ne alter natives i n
Fi gure 5
Decomposition results using resource elements
12
11
10
9
8
7
6
5
4
3
2
1
0
2 cells 3 cells 4 cells 5 cells 6 cells 7 cells
Repeated REs/machines
0.12
0.1
0.08
0.06
0.04
0.02
0
Key
Repeated resource elements
Repeated machines (none)
Validity measure
Validity measure
[ 222 ]
Nabil N.Z. Gindy and
Tsvetan M. Ratchev
Cellular decomposition of
manufacturing facilities using
resource elements
Integrated Manufacturing
Systems
8/ 4 [1997] 215222
advance, thus achi evi ng the effect of multi ple
component routei ngs wi thout the computa-
ti onal complexi ty associ ated wi th them.
Resource elements can be used i n a wi de
range of tasks whi ch requi re capabi li ty based
descri pti on of component requi rements and
machi ni ng faci li ti es (process planni ng, pro-
ducti on planni ng, scheduli ng, etc.).
The reported groupi ng approach based on
resource elements faci li tates more reali sti c
component clusteri ng, generi cally reecti ng
the commonali ty of the component process-
i ng requi rements. By usi ng resource
elements components can be clustered on the
basi s of thei r uni que capabi li ty requi rements
wi thout i t bei ng necessary to si ngle out spe-
ci c machi ne routei ngs among the avai lable
alter natives. The deni ti on of the cell bound-
ari es i n ter ms of resource elements provi des
greater exi bi li ty for deci si on maki ng and
i mplementati on of di fferent company speci c
strategi es i n cell for mati on tasks.
References
1 Tetzlaff, U.A.W., Selecti on of manufacturi ng
equi pment for exi ble producti on systems, i n
Kusi ak, A. (Ed.), I ntelligent Design and Manu-
facturing, J ohn Wi ley & Sons, Chi chester, 1992,
pp. 222-53.
2 Si ngh, N., Desi gn of cellular manufacturi ng
systems: an i nvi ted revi ew, European J ournal
of Operational Research, Vol. 69, pp. 284-91.
3 Gunasekaran, A., Goyal, S.K., Vi rtanen, I . and
Yli -Olli , P., An i nvesti gati on i nto the appli ca-
ti on of group technology i n advanced manufac-
turi ng systems, I nternational J ournal of
Computer I ntegrated Manufacturing, Vol. 7
No. 4, pp. 215-28.
4 Burbri dge, J ., Production Flow Analysis for
Planning Group Technology, Oxford Universi ty
Press, Oxford, 1989.
5 Rajamani , D., Si ngh, N. and Aneja, Y., I nte-
grated desi gn of cellular manufacturi ng sys-
tems i n the presence of alternative process
plans, I nternational J ournal of Production
Research, Vol. 28 No. 8, pp. 1541- 54.
6 Rajamani , D., Si ngh, N. and Aneja, Y., Selec-
ti on of parts and machi nes for cellulari sati on:
a mathemati cal programmi ng approach,
European J ournal of Operational Research,
Vol. 62, pp. 47-54.
7 Kang, S.L. and Wemmerlov, U., Load ori ented
heuri sti c methodology, European J ournal of
Operational Research, Vol. 69, pp. 292-311.
8 Wemmerlov, U. and Hyer, N.L., Cellular manu-
facturi ng i n US i ndustry: a survey of users,
I nternational J ournal of Production Research,
Vol. 27 No. 9, pp. 1511-30.
9 Chu, C.H. and Hayya, J .C., A fuzzy clusteri ng
approach to manufacturi ng cell for mati on,
I nternational J ournal of Production Research,
Vol. 29 No. 7, pp. 1475-87.
10 Xu, H. and Wang, H.-P., Part fami ly for mati on
for GT appli cati ons based on fuzzy mathemat-
i cs, I nternational J ournal of Production
Research, Vol. 27 No. 9, pp. 1637-51.
11 Xi e, X.L. and Beni , G., A vali di ty measure for
fuzzy clusteri ng, I EEE Transactions on Pat-
tern Analysis and Machine I ntelligence, Vol. 13
No. 8, pp. 841-7.
12 Gi ndy, N.N.Z. and Ratchev, T.M., I ntegrated
process capabi li ty model for metal cutti ng and
for mi ng processes, I nternal report, Grant
GR/ G35657-94-1, Loughborough Universi ty of
Technology, 1994.
13 Gi ndy, N.N.Z. and Ratchev, T.M., Product and
machi ne tools data models for computer ai ded
process planni ng systems, i n Domei ngts, G.,
Brown, J . and Tomljanovi ch, M. (Eds),
Computer Applications in Production and
Engineering: I ntegration Aspects, Elsevi er,
Amsterdam, 1991, pp. 527-34.
14 Bezdek, J ., A convergence theorem for the
fuzzy I SODATA clusteri ng algori thms, I EEE
Transactions on Pattern Analysis and Machine
I ntelligence, Vol. 2 No. 1, pp. 1-8.
15 Warren, G. and Moodi e, C., Cellular manufac-
turi ng, Report TAP930104, Purdue Universi ty,
1993.
Fi gure 6
Decomposition results using component-machine formalization
12
11
10
9
8
7
6
5
4
3
2
1
0
2 cells 3 cells 4 cells 5 cells 6 cells 7 cells
Repeated machines
0.12
0.1
0.08
0.06
0.04
0.02
0
Key
Repeated machines
Validity measure
Validity measure
Fi gure 7
Number of machines used for cell determination
40
30
20
10
0
2 cells 3 cells 4 cells 5 cells 6 cells 7 cells
Key
Machine based decomposition
RE based decomposition
32
8
33
12
35
13
34
16
34
18
35
20