Group Technology Implementation in

Manufacturing Systems
TUJUAN INSTRUKSIONAL

• Mahasiswa mampu menyelesaikan

permasalahan tata letak fasilitas
dengan pendekatan Grup Teknologi
Group Technology and Cellular
Manufacturing
• Group technology (GT)
–A management philosophy that
attempts to group products with similar
design or manufacturing characteristics,
or both.
• Cellular manufacturing (CM)
–An application of GT that involves
grouping machines based on the parts
manufactured.
Objective of CM

• Identify machine cells and part families

simultaneously
• Allocate part families to machine cells in
a way that minimizes the intercellular
movement of parts.
CM Concept in Layout

• Develop the layout of machines within

the cells so as to minimize inter- and
intracellular material-handling costs.
Benefits of CM application

• Set-up time reduction

• Work-in-process inventory reduction
• Material-handling cost reduction
• Direct and indirect labor cost reduction
• Improvement in quality
• Improvement in material flow
• Improvement in machine utilization
• Improvement in space utilization
• Improvement in employee morale
A GT Cell
Traditional Job Shop vs CM

• Job shop environment

–Machines are grouped on the basis of
their functional similarities
• CM environment
–Machines are grouped into cells, with
each cell dedicated to the manufacture
of a specific part family
 BM = broaching machine
Arrangement of Cells in a DM = drilling machine
TM = turning machine
Job Shop Environment
VMM = vertical milling machine
Routing of parts P1, P3, P9
Routing of parts P2, P4, P7, P8
Routing of parts P5, P6, P10
 BM = broaching machine
DM = drilling machine
Arrangement of Cells in a TM = turning machine
CM System VMM = vertical milling machine
Sample Part-Machine Processing Indicator
Matrix
Rearranged Processing Indicator
Matrix
Case of Exceptional Parts

• The rows (parts) corresponding to the

1s that lie outside the diagonal block.
• When exceptional parts are removed, a
block diagonal structure is easily
identified.
• If it is wanted that cells are completely
independent with no intercellular
movement of material, exceptional parts
must be subcontracted out
Processing Indicator Matrix (Case
of Exceptional Parts)
Case of Bottleneck Machines

• Bottleneck Machines
–Machines corresponding to the columns that
contain exceptional elements, i.e., elements
outside the block diagonal structure.
–Two or more part families share the machines.
–If the columns corresponding bottleneck
machines are removed, then mutually
separable clusters or machine cells and part
families can be identified.
–Additional copies of machines are needed
Processing Indicator Matrix (Case
of Exceptional Parts)
Using Non-binary in Part-Machine
Processing Indicator Matrix

• Binary matrix representation

–Only informs whether or not a part is
processed on a machine
• Nonbinary matrix representation
–Flexible because it allows to capture
other relationships between each part-
machine pair (e.g., cost of processing a
part on a machine, processing time)
Modifying Part-Machine
Processing Indicator Matrix

• Creating additional columns

–Number of parts to be manufactured
–Batch size for each part
• Sequence of machines visited by a part
can be recorded
–Operation sequence for each part is a
critical factor in identification of machine
cells
Operation Sequence

• Definition of operation sequence

• If the same machine is visited for two or

more nonconsecutive operations:
–Each entry is changed to a vector that
operations for which the corresponding
machine is visited by the part
–Use dummy column
Processing Indicator Matrix
showing Sequence of Operations
Clustering Approach

• Attempt to uncover and display similar

clusters or groups in an input object-object
or object-attribute data matrix.
• Rearrange rows and column of the input
matrix – typically a binary matrix – that
determines whether or not a part is
processed on a particular machine (i.e., a
block diagonal is identified)
• Use process plan or part routing
information
Input Matrix
Common Clustering Algorithms

1.Rank order clustering

2.Bond energy algorithm
3.Row and column masking
4.Similarity coefficient
5.Mathematical programming approach
Rank Order Clustering (ROC)
Algorithm
• ROC algorithm:
–Determine a binary value for each row
and column
–Rearrange the rows and columns in
descending order of their binary values
–Identify clusters.
Steps of ROC Algorithm

• VStep 1:
Assign binary weight BWj = 2m-j to each
column j of the part-machine processing
indicator matrix.

• Step 2:
Determine the decimal equivalent DE of the
binary value of each row i using the formula
Steps of ROC Algorithm

• Step 3:
Rank the rows in decreasing order of their
DE values. Break ties arbitrarily.
Rearrange the rows based on this ranking.
If no rearrangement is necessary, stop;
otherwise go to step 4.

• Step 4:
For each rearranged row of the matrix,
assign binary weight BWi = 2n-i.
Steps of ROC Algorithm
• Step 5:
Determine the decimal equivalent DE of the
binary value of each column j using the formula

• Step 6:
Rank the columns in decreasing order of their
DE values. Break ties arbitrarily. Rearrange the
columns based on this ranking. If no
rearrangement is necessary, stop; otherwise go
to step 1.
 Example – ROC
Part-machine processing indicator matrix
Step 1: Assign binary weight (BW) to each column j of
the part-machine processing indicator matrix
Step 2: Determine the decimal equivalent
(DE) of the binary value for each row i
Step 3: Rank the row in decreasing order of
their DE value and rearrange them based
on this ranking
Step 4
Assign binary weight (BW) to each row i
Step 5: Determine the decimal equivalent (DE) of the
binary value for each column j
Step 6: Rank the column in decreasing order of their
DE value and rearrange them based on this ranking
(Break ties arbitrarily).
Step 1: Assign binary weight (BW) to each
column j
Step 2: Determine the decimal equivalent
(DE) of the binary value for each row i
Step 3: Rank the row in decreasing order of
their DE value and rearrange them based
on this ranking
Step 4: Assign binary weight (BW) to each
row i
Step 5: Determine the decimal equivalent
(DE) of the binary value for each column j
Bond Energy (BE) Algorithm

• Maximize the sum of the bond energies for

each element {i, j} in the part-machine
processing indicator matrix [aij]
• The bond energy is defined so that a
matrix with clusters or diagonal blocks of
1s will have a larger bond energy
compared with the same matrix with rows
and columns arranged so that 1s are
uniformly distributed throughout the matrix.
Bond Energy (BE) Algorithm

• The bond energy for element {i, j} is

given by

• The BEA attempts to maximize the sum

of the bond energies over all row and
column permutations of the part-
machine processing indicator matrix
[aij].
Bond Energy (BE) Algorithm

• By definition, a0,j = an+1,j = ai,0 =

ai,m+1= 0,.
• Rather than maximize the bond energy
in one pass, the BEA performs two
passes, one for row and another for the
column, and each pass it maximizes the
row and column bonds, respectively.
Steps in BEA
• Step 1; ]
Set i = 1. Arbitrarily select any row and place it.
• Step 2:
Place each of the remaining n – 1 rows in each of
the i + 1 positions (i.e., above and below the
previously placed i rows) and determine the row
bond energy for each placement using the formula

Select the row that increase the bond energy the

most and place it in the corresponding position.
• Step 3:
Set i = i + 1. If i < n, go to step 2; otherwise, go to step 4.
• Step 4:
Set j = 1. Arbitrarily select any column and place it.
• Step 5:
Place each of the remaining m – 1 rows in each of the j + 1
positions (i.e., to the left and right of the previously placed j
columns) and determine the column bond energy for each
placement using the formula

Select the column that increase the bond energy the most and
place it in the corresponding position.
• Step 6:
Set j = j + 1. If j < m, go to step 5; otherwise, stop.
Example – Part-machine
processing indicator matrix
Example

• Step 1: Set i = 1. Arbitrarily select row 2

and place it.
• Step 2: Place each of the remaining
three rows in each the two positions and
determine the row bond energy for each
placement
Example

• Step 3: Set i = i + 1 =2. Because i < 4,

go to step 2.
• Step 2: Place each of the remaining two
rows in each of the three positions and
determine the row bond energy for each
placement
Example

• Step 3: Set i = i + 1 = 3. Because i < 4,

go to step 2.
• Step 2: Place the remaining row 4 in
each of the four positions and determine
the row bond energy for each
placement
Example

• Step 3: Set i = i + 1 = 4. Because i = 4,

go to step 4.
• Step 4: Set j = 1. Arbitrarily select
column 1 and place it.
• Step 5: Place each of the remaining
three columns in each of the two
positions and determine the column
bond energy for each placement
Placing column 3 to the left of column 1 maximizes
the column bond energy
Example

• Step 6: Set j = j + 1= 2. Because j < 4,

go to step 5.
• Step 5: Place each of the remaining two
columns in each of the three positions
and determine the column bond energy
for each placement
Example

• Step 6: Set j = j + 1= 3. Because j < 4,

go to step 5.
• Step 5: Place the remaining column in
each of the four positions and determine
the column bond energy for each
placement
Placing column 4 to the right of columns 1, 2
and 3 maximizes the column bond energy
Example

• Step 6: Set j = j + 1= 4. Because j = 4,

stop.
Final clustering:
Row and Column Masking (R&CM)
Algorithm

• Step 1:
Draw a horizontal line through the first
row. Select any 1 entry in the matrix
through which there is only one line.
• Step 2:
If the entry has a horizontal line, go to
step 2a. If the entry has a vertical line
go to step 2b.
Row and Column Masking (R&CM)
Algorithm

• Step 2a:
Draw a vertical line through the column
in which this 1 entry appears. Go to step
2.

• Step 2b:
Draw a horizontal line through the row
in which this 1 entry appears. Go to step
3.
Row and Column Masking (R&CM)
Algorithm
• Step 3:
If there are any 1 entries with only one line
through them, select any one and go to step 2.
Repeat until there are no such entries left.
Identify the corresponding machine cell and
part family. Go to step 4.
• Step 4:
Select any row through which there is no line. If
there are no such rows, stop. Otherwise, draw
a horizontal line through the row, select any 1
entry in the matrix through which there is only
one line, and go to step 2.
Example – Part-machine
processing indicator matrix
Example
Example
Example
Example
Example
Column 1, 4, 6 M1, M4, M6 in cell 1
Row 1, 3 P1, P3 in cell 1
Part Family
Similarity Coefficient (SC)
Algorithm
• SC algorithms are derived from numeric
taxonomy and attempt to measure the similarity
coefficient (SC) between pair of machines or
parts.
• Most of SC algorithms use the Jaccard
similarity coefficient (Sneath and Sola, 1973).
• For a pair of machines, the Jaccard coefficient
is defined as the number of parts that visit both
machines divided by the number of parts that
visit at least one machines
Similarity Coefficient

• The Jaccard coefficient:

where aij = 1 if part k requires processing

on machine i, aij = 0 otherwise.
Example – Part-machine
processing indicator matrix
Using a threshold value
of 0.66, combine
machines
{2, 5} and {4, 6} into
separate cells
SC values in the second iteration

Using a threshold value

of 0.5, combine machines
{1, 4, 6} and {2, 3, 5} into
two cells, respectively
SC values in the third iteration

Using a threshold value

of 0.33, combine machines
{2, 3, 5, 7} into one cell
SC values in the fourth iteration

Using a threshold value

of 0.01, no further combining of cells is
possible
A solution with two cells is obtained;
cell 1 consists of machines 1, 4 and 6.
cell 2 consists of machines 2, 3, 5 and 7
Mathematical Programming
Approach

• A mathematical programming model to

identify part families.
–p-median model
Mathematical Programming
Approach

• The Jaccard coefficient (see below)

measures the (processing) similarity of
parts:

where aij = 1 if part k requires

processing on machine i, aij = 0
otherwise.
Mathematical Programming
Approach
• The weighted Minkowski metric
measures the dissimilarity

where:
r = a positive integer
wk = the weight for part k
Mathematical Programming
Approach

SPECIAL CASE
• If wk = 1 for k = 1, …, n the
Minkowski metric.
• For the Minkowski metric (i.e, wk = 1,
k)
r = 1 the absolute Minkowski metric
r = 2 the Euclidean metric
 p-Median Model
Using the absolute Minkowski metric:
p-Median Model

• P = a parameter that represents the

number of part families desired.
• xij = a decision variable that takes on 0
or 1 integer values only (i.e., xij = 1
indicates that part i belongs to part
family j, xij = 0 otherwise) [Constraint
(5)].
p-Median Model
•
• Constraint (2) ensures that each part belongs
to one part family only.
• Constraint (3) specifies the desired number of
families.
• Constraint (4) guarantees that part i is assigned
to part family j only when this family is formed.
If xjj = 0 for some j, then this part family j is not
formed, so due to constraint (4), no other part i
can be assigned to it.
• The objective function (1) minimizes the overall
dissimilarities of parts.
Example – Part-machine
processing indicator matrix
Dissimilarities Coefficient Matrix
(using the absolute Minkowski matrix)
Sets:
Part/1..6/
ObjFn(PartmPart):D,X)
Endsets
Data:
D= 061556
605132
150445
514023
534203
6 2 5 3 3 0;
p = 2;
Enddata
! Objective function
Min= @SUM(ObjFn(i,j): D*X;
!Constrints
@FOR (Part(i): @SUM(ObjFn(i,j):X(i,j)=1);
@SUM ObjFn(j,j): X(j,j))=P;
@FOR (Part(i):@FOR(Part(j):X(i,j)<=X(j,j)));
@FOR (ObjFn(i,j):@BIN(X));
•Solution: two part families
PF3 = {1, 3}
PF4 = {2, 4, 5, 6}

•From part-machine processing indicator matrix, machine cells can be

identified:

MC1 = {1, 4, 6}
MC2 = {2, 3, 5, 7}
Solution

• Solution: two part families

PF1 = {1, 3}
PF2 = {2, 4, 5, 6}

• From part-machine processing indicator

matrix, machine cells can be identified:
MC1 = {1, 4, 6}
MC2 = {2, 3, 5, 7}