AACIMP 2010 Summer School lecture by Dmitry Krushinsky. "Applied Mathematics" stream. "The p-Median Problem and Its Applications" course. Part 5.
More info at http://summerschool.ssa.org.ua
2. Outline
• Introduction
• Brief overview of models
• PMP approach to CF
• Examples
3. Group technology
• A paradigm in industrial engineering
suggesting structural decomposition of a
manufacturing system into smaller
subsystems.
4. Group technology: advantages
• Smaller systems are easier to manage
(e.g. scheduling)
• Better plant layout:
– shorter in travelling distances (up to 95%)
– less intersecting routes
5. Cell formation (CF)
• Grouping machines into manufacturing
cells …
• … and parts into product families …
• … such that each family is produced
(mainly) within one cell
6. Cell formation
• Cell Formation becomes possible by
exploiting similarities in the manufacturing
processes for different parts and increases
the throughput of the manufacturing
system without sacrificing the products
quality.
19. Performance measures
• ne
• ne + nv
no ne m r no nv
• (1 )
no ne nv m r no nv ne
[0;1] weighting factor
• many others
20. Problem size reduction
machine-part machine-machine
incidence matrix machine-machine similarity matrix
similarity measure
mxr mxm
Similarity measures: r parts
Each machine is characterized by a M1 1 . 1 . . 1
vector in r-dimensional space M2 . 1 . 1 . 1
M3 . . . 1 1 1
similarity any computable metrics M4 1 . 1 . . .
s(i, j ) M5 . 1 . . 1 .
Dissimilarity: d (i, j ) s(i, j ) const
21. Problem size reduction
machine-part machine-machine
incidence matrix machine-machine incidence matrix
similarity measure
mxr mxm
Wei and Kern similarity measure:
r
s(i, j ) (aik , a jk )
r – number of parts,
k 1
m – number of machines
r 1, if aik a jk 1
(aik , a jk ) 1, if aik a jk 0
0, if aik a jk
22. CF: existing approaches
• Clustering based on energy functions:
BEA, ROC, MODROC, DCA, …
• Similarity based hierarchical clustering:
SLC, CLC, ALC, LCC, …
• Fuzzy logic methods
• Genetic algorithms and simulated annealing
• Neural networks:
backpropagation learning, competitive learning, adaptive
resonance theory (ART1), self-organizing maps, …
• Graph partitioning
• Integer Linear Programming
23. Existing approaches: BEA
BEA = bond energy analysis
Goal: minimize the length of the border
1 . 1 . . 1 1 1 1 . . .
. 1 . 1 . 1 1 1 . . . .
. . . 1 1 1 . . 1 1 1 .
1 . 1 . . . . . 1 1 . 1
. 1 . . 1 . . . . . 1 1
• equivalent to the Quadratic Cost Assignment Problem
• only partial solution
24. Existing approaches: hierarchical
clustering
SLC/CLC/ALC = single/complete/average linkage clustering
Algorithm:
• start with each cluster containing one machine
• at each step connect two most similar clusters
25. Existing approaches: hierarchical
clustering
SLC/CLC/ALC = single/complete/average linkage clustering
Algorithm:
• start with each cluster containing one machine
• at each step connect two most similar clusters
26. Existing approaches: hierarchical
clustering
SLC/CLC/ALC = single/complete/average linkage clustering
Algorithm:
• start with each cluster containing one machine
• at each step connect two most similar clusters
27. Existing approaches: hierarchical
clustering
SLC/CLC/ALC = single/complete/average linkage clustering
Algorithm:
• start with each cluster containing one machine
• at each step connect two most similar clusters
28. Existing approaches: hierarchical
clustering
SLC/CLC/ALC = single/complete/average linkage clustering
Algorithm:
• start with each cluster containing one machine
• at each step connect two most similar clusters
Equivalent to
the minimum
spanning tree
problem
(MST)
29. P-Median approach
Goal:
• select p “central” machines – representatives of p
cells
• assign all other machines to cells...
• ... such that the sum of dissimilarities is minimum
30. P-Median approach
Goal:
• select p “central” machines – representatives of p
cells
• assign all other machines to cells...
• ... such that the sum of dissimilarities is minimum
p=2
31. Example 1: functional grouping
Goal: group machines into clusters
Machine-part (manufacturing cells) such as to minimize
incidence matrix intercell communication.
parts
1 2 3 4 5 6 r
1 . 1 . . 1
1
d (i, j ) r (r 1) (aik , a jk )
machines
. 1 . 1 . 1
2
k 1
. . . 1 1 1
3
1 . 1 . . .
4
Wei and Kern’s “commonality score”
. 1 . . 1 .
5
r 1, if aik a jk 1
m = 4, r = 5
(aik , a jk ) 1, if aik a jk 0
0, if aik a jk
r – number of parts, m – number of machines
32. Example 1: functional grouping
Cost matrix for the PMP r
is a machine-machine d (i, j ) r (r 1) (aik , a jk )
dissimilarity matrix:
k 1
cij : d (i, j )
1 . 1 . . 1 12 24 24 17 29
. 1 . 1 . 1 24 12 18 29 23
. . . 1 1 1 24 18 12 29 23
1 . 1 . . . 17 29 29 16 28
. 1 . . 1 . 29 23 23 28 16
33. Example 1: functional grouping
12 24 24 17 29 BC (y ) 12 5 y1 7 y1 y 4 0 y1 y 2 y 4 5 y1 y 2 y 3 y 4
24 12 18 29 23 12 6 y2 5 y 2 y3 1y 2 y3 y5 5 y1 y 2 y 3 y 5
12 6 y3 5 y 2 y3 1y 2 y3 y5 5 y1 y 2 y 3 y 5
C 24 18 12 29 23
16 1y 4 11 y1 y 4 1y1 y 4 y 5 0 y1 y 2 y 4 y 5
17 29 29 16 28
16 7 y5 0 y 2 y5 5 y 2 y3 y5 1y 2 y3 y 4 y5
29 23 23 28 16
BC , p 2 (y) 68 5 y1 6 y2 6 y3 1y 4 7 y5 18 y1 y 4 10 y 2 y3 1y1 y 4 y5 7 y 2 y3 y5
34. Example 1: functional grouping
12 24 24 17 29 BC (y ) 12 5 y1 7 y1 y 4 0 y1 y 2 y 4 5 y1 y 2 y 3 y 4
24 12 18 29 23 12 6 y2 5 y 2 y3 1y 2 y3 y5 5 y1 y 2 y 3 y 5
12 6 y3 5 y 2 y3 1y 2 y3 y5 5 y1 y 2 y 3 y 5
C 24 18 12 29 23
16 1y 4 11 y1 y 4 1y1 y 4 y 5 0 y1 y 2 y 4 y 5
17 29 29 16 28
16 7 y5 0 y 2 y5 5 y 2 y3 y5 1y 2 y3 y 4 y5
29 23 23 28 16
BC , p 2 (y) 68 5 y1 6 y2 6 y3 1y 4 7 y5 18 y1 y 4 10 y 2 y3 1y1 y 4 y5 7 y 2 y3 y5
New variables: Additional constraints:
z6 y1 y 4 z6 y1 y4 1
z7 y 2 y3 z7 y2 y3 1
z8 y1 y 4 y5 z8 y1 y4 y5 2 or z 8 z6 y5 1
z9 y 2 y3 y5 z9 y2 y3 y5 2 or z 9 z7 y5 1
37. Example 2: workforce expenses
Goal: group machines into clusters
Machine-worker (manufacturing cells) such that:
incidence matrix
1) every worker is able to operate every
workers machine in his cell and cost of additional cross-
1 2 3 4 5 6 7 8 training is minimized;
1 0 0 0 1 0 1 0
1
machines
1 1 0 0 0 1 0 0 2) if a worker can operate a machine that is not
2
0 1 1 0 1 0 0 1 in his cell then he can ask for additional
3
0 0 1 1 0 1 0 0 payment for his skills; we would like to minimize
4
0 0 0 1 0 0 1 1 such overpayment.
5
Dissimilarity measure for machines
number of workers that can operate both machines i and j
d (i, j )
number of workers that can operate either of the machines
41. Example 3: from [BhaSad] (2010)*
105 parts
46 machines
(uncapacitated)
functional grouping
105 parts
46 machines
grouping efficiency:
[BhaSad] 90.98%
our result 95.20%
(solved within 1 sec.)
* R. Bhatnagar, V. Saddikuti, Models for cellular manufacturing systems
design: matching processing requirements and operator capabilities,
Journal of the Operational Research Society, 61, 2010, pp. 827-839.
42. Future research directions
• Additional real-life constraints
– capacities
– workload
• Additional real-life factors
– operational sequences
– processing and setup times
44. Conclusions
• An efficient model for CF:
– low computing times
– high quality solutions
• BUT: all models in literature (including our)
are heuristics from the CF perspective
• exact model – MINpCUT