Sec. 18.

4 / Analysis of Cellular Manufacturing 529

Other design applications of group technology involve simplification and stan-

dardization of design parameters such as tolerances, inside radii on corners, chamfer
sizes on outside edges, hole sizes, and thread sizes. These measures simplify design pro-
cedures and reduce part proliferation. Design standardization also pays dividends in
manufacturing by reducing the required number of distinct lathe tool nose radii, drill
sizes, and fastener sizes. There is also a benefit in reducing the amount of data and in-
formation that the company must handle. Fewer part designs, design attributes, tools,
fasteners, and so on mean fewer and simpler design documents, process plans, and other
data records.

18.4 Analysis of Cellular Manufacturing

Many quantitative techniques have been developed to deal with problems in group tech-
nology and cellular manufacturing. Two problem areas are considered in this section:
(1) grouping parts and machines into families, and (2) arranging machines in a GT cell. The
first problem area has been the subject of academic research, and several publications are
listed in references [2], [3], [12], [13], [23], and [24]. The technique described here for solv-
ing the part and machine grouping problem is rank-order clustering [23]. The second prob-
lem area has also been the subject of research, and several reports are listed in references
[1], [7], [9], and [19]. In Section 18.4.2, a heuristic approach by Hollier is introduced [19].

18.4.1 Rank-order Clustering

The problem addressed here is determining how machines in an existing plant should
be grouped into machine cells. The problem is the same whether the cells are virtual or
formal (Section 18.3). It is basically the problem of identifying part families. After part
families have been identified, the machines to produce a given part family can be selected
and grouped together.
The rank-order clustering technique, first proposed by King [23], is specifically
­applicable in production flow analysis. It is an efficient and easy-to-use algorithm for
grouping machines into cells. In a starting part-machine incidence matrix that might be
compiled to document the part routings in a machine shop (or other job shop), the oc-
cupied locations in the matrix are organized in a seemingly random fashion. Rank-order
clustering works by reducing the part-machine incidence matrix to a set of diagonalized
blocks that represent part families and associated machine groups. Starting with the ini-
tial part-machine incidence matrix, the algorithm consists of the following steps:

1. In each row of the matrix, read the series of 1s and 0s 1blank entries = 0s2 from left
to right as a binary number. Rank the rows in order of decreasing value. In case of a
tie, rank the rows in the same order as they appear in the current matrix.
2. Numbering from top to bottom, is the current order of rows the same as the rank order
determined in the previous step? If yes, go to step 7. If no, go to the following step.
3. Reorder the rows in the part-machine incidence matrix by listing them in decreasing
rank order, starting from the top.
4. In each column of the matrix, read the series of 1s and 0s 1blank entries = 0s2
from top to bottom as a binary number. Rank the columns in order of decreasing
value. In case of a tie, rank the columns in the same order as they appear in the cur-
rent matrix.

M18_GROO6119_04_GE_C18.indd 529 2/28/15 4:13 PM

 530 Chap. 18 / Group Technology and Cellular Manufacturing

5. Numbering from left to right, is the current order of columns the same as the rank order
determined in the previous step? If yes, go to step 7. If no, go to the following step.
6. Reorder the columns in the part-machine incidence matrix by listing them in de-
creasing rank order, starting with the left column. Go to step 1.
7. Stop.

For readers unaccustomed to evaluating binary numbers in steps 1 and 4, it might be

helpful to convert each binary value into its decimal equivalent. For example, the entries
in the first row of the matrix in Table 18.3 are read as 100100010. This converts to its
decimal equivalent as follows:11 * 28 2 + 10 * 27 2 + 10 * 26 2 + 11 * 25 2 + 10 * 24 2
+ 10 * 23 2 + 10 * 22 2 + 11 * 21 2 + 10 * 20 2 = 256 + 32 + 2 = 290. Decimal con-
version becomes impractical for the large numbers of parts found in practice, so it is pref-
erable to compare the binary numbers.
In Example 18.1, it is possible to divide the parts and machines into three mutually
exclusive part-machine groups. This represents the ideal case because the part families and

Example 18.1 Rank-order Clustering Technique

Apply the rank-order clustering technique to the part-machine incidence ma-
trix in Table 18.3.
Solution: Step 1 consists of reading the series of 1s and 0s in each row as a binary number.
This is done in Table 18.6(a), converting the binary value for each row to its
decimal equivalent. The values are then rank-ordered in the far right-hand
column. In step 2, it is seen that the row order is different from the starting
matrix. Therefore, the rows are reordered in step 3. In step 4, the series of 1s and
0s in each column are read from top to bottom as a binary number (again this
has been converted to the decimal equivalent) and rank the columns in order
of decreasing value, as shown in Table 18.6(b). In step 5, it is observed that the
column order is different from the preceding matrix. Proceeding from step 6
back to steps 1 and 2, and a reordering of the columns provides a row order that
is in descending value and the algorithm is concluded (step 7). The final solution
is shown in Table 18.6(c). A comparison of this solution with Table 18.4 reveals
that they are the same part-machine groupings.

Table 18.6(a)  First Iteration (Step 1) in the Rank-order Clustering

Technique Applied to Example 18.1
Binary values 28 27 26 25 24 23 22 21 20
Decimal
Parts Equivalent
Machines A B C D E F G H I Rank
1 1 1 1 290 1
2 1 1 17 7
3 1 1 1 81 5
4 1 1 136 4
5 1 1 258 2
6 1 1 65 6
7 1 1 1 140 3

M18_GROO6119_04_GE_C18.indd 530 2/28/15 4:13 PM

 Sec. 18.4 / Analysis of Cellular Manufacturing 531

Table 18.6(b)  Second Iteration (Steps 3 and 4) in the Rank-order

Clustering Technique Applied to Example 18.1
Parts

Binary
Machines A B C D E F G H I values
1 1 1 1 26
5 1 1 25
7 1 1 1 24
4 1 1 23
3 1 1 1 22
6 1 1 21
2 1 1 20
Decimal 96 24 6 64 5 24 16 96 7
equivalent Rank 1 4 8 3 9 5 6 2 7

Table 18.6(c)  Solution of Example 18.1

Parts

Machines A H D B F G I C E
1 1 1 1
5 1 1
7 1 1 1
4 1 1
3 1 1 1
6 1 1
2 1 1

associated machine cells are completely segregated. However, it is not uncommon for an
overlap in processing requirements to exist between machine groups. That is, a given part
type needs to be processed by more than one machine group. One way of dealing with the
overlap is simply to duplicate the machine that is used by more than one part family, plac-
ing the same machine type in both cells. Other approaches, attributed to Burbidge [23],
include (1) change the routing so that all processing can be accomplished in the primary
machine group, (2) redesign the part to eliminate the processing requirement outside the
primary machine group, and (3) purchase the parts from an outside supplier.

18.4.2 Arranging Machines in a GT Cell

After part-machine groupings have been identified, the next problem is to organize the
machines into the most logical sequence. A simple yet effective method is suggested by
Hollier [19]3 that uses data contained in from–to charts (Section 10.3.1) and is intended
to place the machines in an order that maximizes the proportion of in-sequence moves

3
Hollier [19] presented six heuristic approaches to solving the machine arrangement problem, of which
only one is described here. He presents a comparison of the six methods in his paper.

M18_GROO6119_04_GE_C18.indd 531 2/28/15 4:13 PM