LAYOUT PLANNING MODELS

6
AND DESIGN ALGORITHMS

6.1 INTRODUCTION
The generation and evaluation of a number of layout alternatives is a critical step in
the facilities planning process, since the layout selected will serve to establish the
material flow patterns and physical relationships between activities. Recognizing
that the layout ultimately selected will be either chosen from or based on one of the
alternatives generated, it is important for the facilities planner to be both creative
and comprehensive in generating a reasonable number of layout alternatives.
Most facility layouts can be viewed at two levels: The block layout (which
shows the location, shape, and size of each planning department) and the detailed
layout (which shows the exact location of all equipment, work benches, and stor-
age areas within each department). The block layout is concerned primarily with
the “macro” flows in the facility, while the detailed layout is often concerned with
the “micro” flows. For the facility layout to be complete and effective, both the
block layout and the detailed layout need to be developed and evaluated carefully.
Although some of the models and techniques we show in this chapter can be ap-
plied to detailed layouts as well, our focus is primarily on quantitative methods for
developing and evaluating alternative block layouts.
We previously addressed how to determine the requirements. In Chapter 1, we
treated the strategic relationship between facilities planning and manufacturing, distri-
bution, and marketing. From that discussion, we recognized the importance of taking
a long-range viewpoint and coordinating the facilities plan with the plans of other or-
ganizational units. A facilities layout strategy should emerge from the overall strategic
292
 293
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

plan. Product, manufacturing, marketing, distribution, management, and human re-

source plans will be impacted by and will have an impact on the facilities layout.
The facilities requirements resulting from product design, process design, and
schedule design decisions were examined in Chapter 2. The impact of personnel
requirements on space, proximities, and special features of the facility was treated
in Chapter 4.
Chapter 3 provided a comprehensive treatment of activity relationships, space
requirements, and types of planning departments as they relate to facilities plan-
ning. From that discussion, as well as the emphasis in Chapter 1 on the establish-
ment of activity relationships and determination of space requirements, it is clear
that both the block layout and the detailed layout are critically important in design-
ing a layout for a facility.
Before proceeding further, it seems appropriate to address two questions,
which frequently come up in a layout planning project:
1. Which comes first, the material handling system or the facility layout?
2. Which comes first, the block layout or the detailed layout?
Many appear to believe the layout should be designed first and then that the
material handling system should be developed. Yet material handling decisions can
have a significant impact on the effectiveness of a layout. For example, the follow-
ing decisions will affect the layout:
1. Centralized versus decentralized storage of work-in-process (WIP), tooling,
and supplies
2. Fixed-path versus variable-path material handling
3. The handling unit (unit load) planned for the systems versus one-piece flow
4. The degree of automation used in handling
5. The type or level of inventory control, physical control, and computer control
of materials
Each of the above considerations affects the requirements for space, equipment, and
personnel, as well as the degree of proximity required between various departments.
Why do people tend to focus first on layout? Perhaps one reason is an
overemphasis on the manufacturing process. For example, it seems perfectly logical
to place department B next to department A “if process B occurs immediately after
process A.” In such a situation, the handling problem is reduced to the question,
“What is the best way to move materials from A to B?” Conventional wisdom sug-
gests that the handling problem can be addressed after the layout is finalized.
However, if materials cannot flow directly from department A to department B,
then WIP storage is required in A, B, and/or elsewhere. Depending on the storage
and control requirements, a centralized WIP storage area might be used, so that
materials flow from A to S (storage), and then from S to B. With such a centralized
WIP storage system, materials do not flow from A to B, and B no longer needs to be
placed next to A. Furthermore, the centralized system provides added flexibility
when the process sequence changes. Yet lean manufacturing teaches us to redefine
the departments and create “cells” so that the parts flow, one at a time, directly from
one workstation to the next, which eliminates intermediate storage areas and signif-
icantly reduces the material handling distances and delays incurred due to batching.
 294
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

Another reason for the “layout first” approach could be a misapplication of the
“handling less is best” adage. For example, in one-piece flow (the make one, move
one principle practiced in lean manufacturing), the parts are handled more times
compared to batching (when all the parts are placed in one container and moved
only once as a unit load). Yet, due to delays introduced by batching, in a large ma-
jority of cases, one-piece flow is the preferred approach, provided the workstations
are placed next to each other. Hence, when one says “handling less is best,” one
needs to be careful in how handling is defined and how less or more handling im-
pacts measures such as WIP inventory, the expected time it takes for the parts to
travel through the system, and walk times/material handling times for the operators.
So, which comes first, the material handling system or the facility layout? Our
answer is, “Both!” The layout and the handling system should be designed simulta-
neously. The complexity of the design problem, however, generally requires that a
sequential process be used. For this reason, we recommend that a number of alter-
native layout plans be developed and the appropriate handling system be designed
(at least at a conceptual level) for each alternative. The preferred layout will be that
which results from a consideration of the system as a whole [72].
As for the second question, our recommendation is to first obtain basic require-
ments for each department (such as space requirements, shape constraints, and so
on) and then develop a set of alternative block layouts. Once the desirable block lay-
out is identified, the analyst can then develop a detailed layout for each department.
In the process, he/she may very well go back and “massage” or “modify” the block
layout, creating an iterative process, which can be repeated for some of the other
block layout alternatives as well. Since the process often ends up being an iterative
one, the starting point is less critical, although gathering basic information for each
department and then starting with macro-flows is often a practical approach.

6.2 BASIC LAYOUT TYPES

In Chapter 3, we identified four types of planning departments:
1. Fixed material location departments
2. Production line departments
3. Product family departments
4. Process departments
Based on the above types of departments, we also identified four types of layouts, as
shown earlier in Figure 3.37 and Table 3.2. (For the reader’s convenience, Figure 3.37
is repeated here as Figure 6.1.)
Before we explain in the next section various procedures developed for gener-
ating and evaluating layout alternatives, we need to stress that identifying the right
number and types of departments that are going to be used to populate the layout is
a key step the facilities planner must perform first. A critical document for this pur-
pose is the future state map (FSM). The facilities planner must establish very clearly
the relationship between each “process box” in the FSM and each “department” that
is to appear in the layout. Skipping this step (or performing it with a poorly defined
 Lathe Drill Grind Drill
W
A A
S S
T R
Press Bend Drill S E
O E
R H
M O
E Mill Drill B
S U
L S
Y E
Lathe Lathe Drill

(a)

Lathe Press Grind W

S A
T R
O E
R H
E O
S U
S
Weld Paint Assembly E

(b)

Lathe Drill Grind Assembly

W
Mill Assembly Weld Paint A
S R
T E
O H
R O
E U
Press Lathe Drill Press Assembly S
S
E

Grind Drill Assembly Drill Grind

(c)

Lathe Lathe Drill Weld Weld

W
S A
Lathe Lathe Drill Paint Paint R
T
O E
R H
E O
S Mill Mill Grind Assembly U
S
E
Mill Mill Grind Assembly

(d)

Figure 6.1 Alternative types of layouts. (a) Production line product layout. (b) Fixed
product layout. (c) Product family layout. (d ) Process layout.

FSM) may result in a layout that minimizes the flow distances between the “wrong”
departments; that is, the flow distances in the layout may be minimized, but the de-
partments that populate the layout may not be the right type of departments needed
to support lean manufacturing. (For an introduction to value stream mapping and
basic lean principles used to develop an FSM, the reader may refer to [54].)
295
 296
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

6.3 LAYOUT PROCEDURES

A number of different procedures have been developed to aid the facilities planner in
developing layout alternatives. These procedures can be classified into two main cate-
gories: construction type and improvement type. Construction layout methods basically
involve developing a new layout “from scratch.” Improvement procedures, on the other
hand, generate layout alternatives by seeking improvements in an existing layout.
Although a number of papers in the facility layout literature focus on the de-
velopment of construction-type procedures, most layout work still involves some
form of improving the layout of an existing facility. As Immer [29, p. 32] observed as
early as the 1950s,
Much of [the] work will consist of making minor changes in [an] existing layout,
locating new machines, revising a section of the plant, and making occasional studies
for material handling. The plans for a complete new production line or a new factory
may make the headlines, but except for a war or a new expansion, the average layout
[planner] will very seldom have to consider such a problem.

While Immer’s observation still holds today for the most part, some changes
to the layout are no longer “minor” changes (due to the far-reaching impact lean
manufacturing has had in the manufacturing sector), and there are new plants
being built around the globe, as manufacturing spreads from traditional bases such
as the United States to other regions of the world.
We begin our discussion of layout procedures by discussing some of the orig-
inal approaches to the layout problem. The concepts in these approaches continue
to serve as the foundation of many of the methodologies proposed today.

6.3.1 Apple’s Plant Layout Procedure

Apple [2] proposed the following detailed sequence of steps in producing a plant
layout.
1. Procure the basic data.
2. Analyze the basic data.
3. Design the productive process.
4. Plan the material flow pattern.
5. Consider the general material handling plan.
6. Calculate equipment requirements.
7. Plan individual workstations.
8. Select specific material handling equipment.
9. Coordinate groups of related operations.
10. Design activity interrelationships.
11. Determine storage requirements.
12. Plan service and auxiliary activities.
13. Determine space requirements.
14. Allocate activities to total space.
15. Consider building types.
 297
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

16. Construct master layout.

17. Evaluate, adjust, and check the layout with the appropriate persons.
18. Obtain approvals.
19. Install the layout.
20. Follow up on implementation of the layout.
Apple noted that these steps are not necessarily performed in the sequence given.
As he put it,
Since no two layout design projects are the same, neither are the procedures for de-
signing them. And, there will always be a considerable amount of jumping around
among the steps, before it is possible to complete an earlier one under consideration.
Likewise, there will be some backtracking going back to a step already done—to re-
check or possibly re-do a portion, because of a development not foreseen [2, p. 14].

6.3.2 Reed’s Plant Layout Procedure

Reed [53] recommended the following “systematic plan of attack” as required steps
in “planning for and preparing the layout.”
1. Analyze the product or products to be produced.
2. Determine the process required to manufacture the product.
3. Prepare layout planning charts.
4. Determine workstations.
5. Analyze storage area requirements.
6. Establish minimum aisle widths.
7. Establish office requirements.
8. Consider personnel facilities and services.
9. Survey plant services.
10. Provide for future expansion.
Reed calls the layout planning chart “the most important single phase of the entire
layout process” [53, p. 10]. It incorporates the following:
1. Flow process, including operations, transportation, storage, and inspections
2. Standard times for each operation
3. Machine selection and balance
4. Manpower selection and balance
5. Material handling requirements
An example of a layout planning chart is given in Figure 6.2. Such charts can be viewed
as the predecessors of value stream maps used in lean manufacturing today [54].

6.3.3 Muther’s Systematic Layout Planning (SLP) Procedure

Muther [49] developed a layout procedure he named systematic layout planning, or
SLP. The framework for SLP is given in Figure 6.3. It uses as its foundation the activ-
ity relationship chart described in Chapter 3 and illustrated in Figure 6.4.
298

LAYOUT PLANNING CHART

PART NO. 1 PART NAME PLASTIC CONTACT PAD PCS/ASSY 1 PCS REQ/HR 70.4 SHEET 1 OF 1
ASSY NO. — ASSY NAME — ASSY/PRODUCT 2 PRODUCTION HRS/DAY 6.0 PREPARED BY J.G.D. DATE 1-4-60
MATERIAL PLASTIC SIZE 11/2" OD × 3/8" ID (from 4' × 8' × 1/4" SHEETS) PCS/DAY 422 LOT SIZE 1 APPROVED BY DATE

TIME MACHINE MACHINES REQD OPER TOTAL MANPOWER HANDLING REQUIREMENTS

ST OPER DEPT
FMSI DESCRIPTION PER OR MACH COMB MACH PER CREQ MAN COMB MEN HOW CONT LOAD DIST REMARKS
NO. NO. NO.
PIECE EQUIPMENT FRAC WITH REQD MACH FRAC FRAC WITH REQD MOVED TYPE SIZE MOVED
FORK
1 FROM MATERIALS STORAGE LIFT
PALLET 4 SHEETS 100'

2 ON PALLET BY SAW 2

3 TO SAW TABLE
9 – 10 9 – 10
4 SAW INTO STRIPS 2 1/2" × 8' 10 2 .02 TABLE SAW .028 16 – 10 1 2 .028 .056 16 – 10 2
1 – 20
5 TO RACK BY SAW 14 – 30
17 – 20
6 IN RACK 2 3 – 20
6 – 30
7 TO HEATER 13 – 20
8 IN RACK BY HEATER 2 14 – 20

9 FEED INTO HEATER

9 – 20 6 – 20
10 HEAT 10 2 .04 HEATER .055 16 – 20 1 1 .055 .055 6 – 10 1
3 – 10
2 – 10
9 – 20
16 – 20
13 – 10
14 – 10
17 – 10

11 FEED TO PUNCH PRESS

4 – 10 4 – 10
12 PUNCH TO SHAPE 10 2 .04 PUNCH PRESS .055 4 – 20 1 1 .055 .055 4 – 20 1
9 – 30 9 – 30
13 TP BIN BY PUNCH PRESS 16 – 30 16 – 30
19A – 10 19A – 10
14 IN BIN 2 19B – 10 19B – 10
19A – 20 4 WHEEL TOTE 400
15 TO PARTS STORAGE 5A – 10 HAND TRUCK BOX PER BOX 150'

16 IN PARTS STORAGE 5
4 WHEEL TOTE 400
17 TO ASSEMBLY HAND TRUCK BOX PER BOX 60'

18 IN BIN IN ASSEMBLY 6

19 TO TABLE

Figure 6.2 Layout planning chart.

 299
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

Input data and activities

1. Flow of 2. Activity
materials relationships

Analysis
3. Relationship
diagram

4. Space 5. Space
requirements available

6. Space
relationship
diagram

7. Modifying 8. Practical Search

considerations limitations

9. Develop
layout
alternatives
Selection

10. Evaluation

Figure 6.3 Systematic layout planning (SLP) procedure.

Based on the input data and an understanding of the roles and relationships
between activities, a material flow analysis (from-to chart) and an activity relation-
ship analysis (activity relationship chart) are performed. From the analyses per-
formed, a relationship diagram is developed (Figure 6.5).
The relationship diagram positions activities spatially. Proximities are typically
used to reflect the relationship between pairs of activities. Although the relationship
diagram is normally two-dimensional, there have been instances in which three-
dimensional diagrams have been developed when multistory buildings, mezza-
nines, and/or overhead space were being considered.
 300
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

ACTIVITY RELATIONSHIP CHART Plant TRESISA Project A-35

Charted by J T With
Date 1/14 Sheet 1 of 1
Reference 35

1
1
Receiving

2
A

3
2 1 A This block shows relation
Shipping 1
between "1" and "3"

4
E 2 E

5
3 Raw materials and 4 A 4 O 2 Importance of
parts storage relationship (top)

6
E 2 U U

7
4 Finished goods 4 A O U 3 Reasons in
code (below)

8
storage E 2 U U A

9
5 2 O U E 6 O
Manufacturing

10
A A E 6 U

11
6 Work-in-process 2 A 2 E 4,5
A

12
storage A 2 O 4,5 U 9

13
7 2 O A
Assembly

14
X O 3,4

15
8 8 A
Offices

16
X 3,4

17
9 8
Maintenance

18
19
10

20
11

1
2
12 3
4

13 Value CLOSENESS
5

Absolutely
A Necessary
6

14 Especially
E Important
7

“Closeness” I Important
8

Ordinary
15 rating O Closeness OK
9

U Unimportant
10

16 X Unimportant
11

Code REASON
12

17 1 Same deck
13

2 Flow of material
14

18 3 Service
15

Reasons 4 Convenience
16

behind the 5
19 Inventory control
“closeness”
17

value 6 Communication
18

20 7 Same personnel
19

8 Cleanliness
20

9 Flow of parts

Figure 6.4 Activity relationship chart.

The next two steps involve the determination of the amount of space to be as-
signed to each activity. From Chapter 3, departmental service and area requirement
sheets would be completed for each planning department. Once the space assign-
ments have been made, space templates are developed for each planning department,
and the space is “hung on the relationship diagram” to obtain the space relationship
diagram (Figure 6.6).
 301
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

8 5

3 6

9
2

4 7

Figure 6.5 Relationship diagram.

8 5
(9000) (12,000)

3 6
(9000) (1000)

1
(1000)
9
(1000)

2
(1000)
7
(4000)
4
(9000)

Figure 6.6 Space relationship diagram.

 302
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

1
3
1 2
3
2
9
8 5 7
9
4
4 7
1 9
6 5
3 6 5
2 6

4 7 8 8

(a) (b) (c)

Figure 6.7 Alternative block layouts.

Based on modifying considerations and practical limitations, a number of layout

alternatives are developed (Figure 6.7) and evaluated. The preferred alternative is
then identified and recommended.
While the process involved in performing SLP is relatively straightforward, it
does not necessarily follow that difficulties do not arise in its application. We ad-
dressed in Chapter 3 such issues as the a priori assignment of activity relationships
and the use of proximity as a criterion for measuring the degree of satisfaction of ac-
tivity relationships. In addition to those concerns, it should be noted that alternative
relationship diagrams can often be developed, with apparent equivalent satisfaction
of activity relationships. Likewise, the shapes of the individual space templates used
in constructing the space relationship diagram can influence the generation of alter-
natives. Finally, the conversion of a space relationship diagram into several feasible
layout alternatives is not a mechanical process: intuition, judgment, and experience
are important ingredients in the process.
The SLP procedure can be used sequentially to develop first a block layout
and then a detailed layout for each planning department. In the latter application,
relationships between machines, workstations, storage locations, and entrances to
and exits from the department are used to determine the relative location of activi-
ties within each department.

6.4 ALGORITHMIC APPROACHES

The relative placement of departments on the basis of their “closeness ratings” or
“material flow intensities” is one that can be reduced to an algorithmic process. The
layout procedures we presented in the previous section provide an excellent frame-
work and overall process to construct or improve a layout, but they do not provide
a formal procedure or algorithm for some of the critical steps associated with layout
design and evaluation. The models and algorithms we present in this section are
formal procedures that can help the layout analyst develop or improve a layout and
at the same time provide him or her with objective criteria to facilitate the evalua-
tion of various layout alternatives that emerge in the process.
 303
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

The layout algorithms we present, at least in theory, can be executed by

hand. However, for most practical, real-world problems, the algorithms we show
here are intended or best-suited for computer implementation. Currently available
computer-based layout algorithms cannot replace human judgment and experience,
and they generally do not capture the qualitative characteristics of a layout. However,
computerized layout algorithms can significantly enhance the productivity of the lay-
out planner and the quality of the final solution by generating and numerically eval-
uating a large number of layout alternatives in a very short time. Computerized
layout algorithms are also very effective in rapidly performing “what-if” analyses
based on varying the input data or the layout itself.
The algorithms presented in this section, for the most part, represent the out-
growth of university research. As such, commercial versions of the algorithms we pre-
sent either do not exist or they must be obtained through the original authors. Also,
due to limited space, we are only able to present what we hope is a diverse and rep-
resentative subset of the many layout algorithms that have been developed to date.
A number of commercial packages are available for facility layout. However,
with some exceptions (see Section 6.8), such packages are either intended only for
presentation purposes (that is, they are electronic drafting tools that facilitate the
drawing or maintenance of a given layout) or they are designed primarily as a lay-
out evaluation tool (that is, they can evaluate a layout, provided that one has been
supplied by the layout planner). While such tools can also significantly enhance the
productivity of the layout planner, in this section we will focus on those algorithms
that can actually improve a given layout or develop a new layout from scratch.

6.4.1 Algorithm Classification

Most layout algorithms can be classified according to the type of input data they
require. Some algorithms accept only qualitative “flow” data (such as a relationship
chart), while others work with a (quantitative) flow matrix expressed as a from-to
chart. Some algorithms (such as BLOCPLAN) accept both a relationship chart and a
from-to chart; however, the charts are used only one at a time when evaluating a
layout. The current trend appears to be toward algorithms that use a from-to chart,
which generally requires more time and effort to prepare but provides more infor-
mation on parts flow (or material handling trips) when completed. Given that flow
values can be converted to relationship ratings and vice versa, most algorithms can
be used with either type of data. Of course, if a relationship chart is converted to a
from-to chart (by assigning numerical values to the closeness ratings), then the
“flow” values picked by the layout planner represent only an ordinal scale.
With respect to data input, the time and effort required to compile a relation-
ship chart or a from-to chart increases rapidly with the number of departments.
Since the relationship chart is based on user-assigned closeness ratings, its construc-
tion often requires input from multiple individuals. After obtaining such input, the
analyst needs to identify and resolve possible inconsistencies. For example, for the
same pair of departments, one person may assign an “A,” while another assigns an
“I.” Rather than attempt to decide who is “right” and who is “wrong,” the analyst
needs to meet with both individuals to understand the reasons behind their ratings
and, with agreement from all concerned parties, determine the final closeness rating
 304
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

to be used for the pair. Given the significant increase in the number of all possible
department pairs as a function of the number of departments, relationship charts are
often not practical for problems with 20 or more departments.
The same increase in the number of department pairs also applies to from-to
charts. For medium- to large-sized problems (say, 20–30 or more departments), fill-
ing out each entry in the from-to chart would not be practical. In such cases, how-
ever, the from-to chart can be constructed in reasonable time by using the
production route data for each product (or product family). For example, if product
type A is processed through departments 1-2-5-7, and is moved at a rate of, say, 20
loads per hour, then we set f12  f25  f57  20 in the from-to chart. Repeating this
process for each product (or product type) completes the from-to chart, which is
often a “sparse matrix” (i.e., it contains many blank entries). In fact, in many cases, it
is useful to first construct a separate and complete from-to chart for each product (or
product type) so product-level flow data remain available to the analyst at all times.
Subsequently, individual product-level from-to charts can be combined into one
cumulative chart, using appropriate weights for individual product types if necessary.
Of course, the construction of the from-to chart in the above manner is done by
computers in most cases. Also, if the unit of flow for the product changes as it moves
from one process to the next, then appropriate multipliers can be inserted into the
routing data to scale the flow intensity up or down based on the unit of flow.
Layout algorithms can also be classified according to their objective functions.
There are two basic objectives: one aims at minimizing the sum of flows times dis-
tances, while the other aims at maximizing an adjacency score. Generally speaking,
the former, that is, the “distance-based” objective—which is similar to the classical
quadratic assignment problem (QAP) objective—is more suitable when the input
data is expressed as a from-to chart, and the latter, that is, the “adjacency-based”
objective, is more suitable for a relationship chart.
Consider first the distance-based objective. Let m denote the number of depart-
ments, fij denote the flow from department i to department j (expressed in number
of unit loads moved per unit time), and cij denote the cost of moving a unit load
one distance unit from department i to department j. The objective is to minimize
the cost per unit time for movement among the departments. Expressed mathemat-
ically, the objective can be written as
m m
min z = a a fij cij dij (6.1)
i=1 j=1

where dij is the distance from department i to j. In many layout algorithms, dij is
measured rectilinearly between department centroids; however, it can also be mea-
sured according to a particular aisle structure (if one is specified).
Note that the cij values in Equation 6.1 are implicitly assumed to be independ-
ent of the utilization of the handling equipment, and they are linearly related to the
length of the move. In those cases where the cij values do not satisfy the above as-
sumptions, one may set cij  1 for all i and j and focus only on total unit load travel
in the facility (i.e., the product of the fij and the dij values). In some cases, it may
also be possible to use the cij values as relative “weights” (based on unit load attri-
butes such as size, weight, bulkiness, etc.) and minimize the weighted sum of unit
load travel in the facility.
 305
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

Consider next the adjacency-based objective, where the adjacency score is

computed as the sum of all the flow values (or relationship values) between those
departments that are adjacent in the layout. Letting xij  1 if departments i and j are
adjacent (that is, if they share a border) in the layout, and 0 otherwise, the objective
is to maximize the adjacency score; that is,
m m
max z = a a fij xij (6.2)
i=1 j=1

Although the adjacency score obtained from Equation 6.2 is helpful in comparing
two or more alternative layouts, it is often desirable to evaluate the relative efficiency
of a particular layout with respect to a certain lower or upper bound. For this pur-
pose, the layout planner may use the following “normalized” adjacency score:
m m
a i = 1 a j = 1 fij xij
z = m m (6.3)
a i = 1 a j = 1 fij
Note that the normalized adjacency score (which is also known as the efficiency
rating) is obtained simply by dividing the adjacency score obtained from Equation 6.2
by the total flow in the facility. As a result, the normalized adjacency score is always
between zero and 1. If the normalized adjacency score is equal to 1, it implies that all
department pairs with positive flow between them are adjacent in the layout.
In some cases, the layout planner may represent an X relationship between
departments i and j by assigning a negative value to fij. The exact negative value to
be used should be determined with respect to the “real” (i.e., positive) flow values
in the from-to chart. If such “negative flow” values are used, then the normalized
adjacency score must be modified as follows:

a (i,j )僆F fij xij + a (i,j)僆F fij (1 - xij)

z = (6.4)
a (i,j )僆F fij - a (i,j)僆F fij
where F and F represent the set of department pairs with positive and negative flow
values, respectively.
The adjacency-based objective has been used in a number of algorithms; see, for
example, [15], [16], [17], and [47], among others. Although such an objective is easy to
use and intuitive (i.e., department pairs with high closeness ratings or large flows need
to be adjacent in the layout), it is generally not a complete measure of layout efficiency
since it disregards the distance or separation between nonadjacent departments. There-
fore, as remarked in [6], it is possible to construct two layouts that have identical or
similar adjacency scores but different travel distances in terms of parts flow.
Layout algorithms can further be classified according to the format they use for
layout representation. Most layout algorithms use the discrete representation
(shown in Figure 6.8a), which allows the computer to store and manipulate the lay-
out as a matrix. With such a representation, the area of each department is rounded
off to the nearest integer number of grids. If the grid size is too large (too small),
small (large) departments may have too few (too many) grids. Also, the grid size
determines the overall “resolution” of the layout; a smaller grid size yields a finer
resolution, which allows more flexibility in department shapes. However, since the
 306
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

(a) (b)

Figure 6.8 Discrete versus continuous layout representation. (Adapted from [18].)

total area is fixed, a smaller grid size results in a larger number of grids (or a
larger matrix), which can considerably increase the computational burden. Hence,
in algorithms that use the discrete representation, selecting the appropriate grid size
is an important decision that must be made early in the planning process.
The alternative representation (see Figure 6.8b) is the continuous representation
where there is no underlying grid structure. (The grid in Figure 6.8b was retained only
for comparison purposes.) Although such a representation is theoretically more flexi-
ble than its discrete counterpart, it is also more difficult to implement on a computer.
In fact, except for one case (where the shapes of certain departments are “adjusted”
to accommodate a nonrectangular building), present computerized layout algorithms
that use the continuous representation are restricted to a rectangular building and rec-
tangular department shapes.
While it is possible to model nonrectangular buildings by using fixed
“dummy” departments (see the section on CRAFT), generally speaking, defining
L-shaped, U-shaped, and other arbitrary, nonrectangular departments with the con-
tinuous representation is not straightforward. If a department is rectangular, and we
know its area, then we need to know only the x, y-coordinate of its centroid and the
length of its side in the north–south direction to specify its exact location and shape.
(How would you specify the exact location and shape of an L-shaped department
or a T-shaped department with a known area? Among alternative specifications,
which one requires the minimum amount of data? Which one makes it easier to
identify and avoid overlapping departments?)
Department shapes play an important role in computerized layout algorithms.
Although we discuss specific shape measures in Section 6.5, first we need to pre-
sent some basics. Since, by definition, a department represents the smallest indivisi-
ble entity in layout planning, a layout algorithm should not “split” a department into
two or more pieces. If some departments are “too large,” the layout planner can go
back and reconsider how the departments were defined and change some of them
as necessary. In the process, one large department may be redefined as two smaller
departments. Once all the departments have been defined, however, a layout algo-
rithm cannot change them or split them.
Although the human eye is adept at judging shapes and readily identifying
split departments, for the computer to “recognize” a split department, we need to
develop formal measures that can be incorporated into an algorithm. Consider, for
example, the discrete representation, where a department is represented as a collec-
tion of grids. Suppose there is a “dot” that can move only from one grid to an adja-
cent grid. (Two grids are adjacent only if they share a border of positive length; two
 307
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

(a) (b) (c)

(d) (e)
Figure 6.9 Examples of split and unsplit departments.

grids that “touch” each other at the corners are not considered adjacent.) We say
that department i is not split if the above dot can start at any grid assigned to depart-
ment i and travel to any other grid assigned to department i without visiting any grid
that has not been assigned to department i. In other words, given the restrictions
imposed on the movements of the dot, any grid assigned to department i must be
“reachable” from any other such grid.
For example, the department shown in Figure 6.9a and b is split, while those
shown in Figure 6.9c and d are not. However, according to the above definition, the
department shown in Figure 6.9e is also not a split department. Departments such as
the one shown in Figure 6.9e are said to contain an “enclosed void” [18] and as a rule
of thumb are not considered practical or reasonable for facility layout purposes. (Of
course, one possible exception to this rule is an “atrium.” However, an atrium should
generally be modeled by placing a fixed “dummy department” inside the building. Re-
fer to the section on CRAFT for dummy departments.) The shape measures we present
in Section 6.5 are devised to generally avoid department shapes such as those shown
in Figure 6.9d and e. In the above discussion of department shapes, we focused on the
discrete representation. Although it has been used effectively only with rectangular
departments, our comments apply to the continuous representation as well.
Lastly, as we remarked earlier, layout algorithms can be classified according
to their primary function—that is, layout improvement versus layout construction.
Improvement-type algorithms generally start with an initial layout supplied by the
analyst and seek to improve the objective function through “incremental” changes
in the layout. Construction-type layout algorithms generally develop a layout “from
scratch.” They can be further divided into two categories: those that assume the
building dimensions are given and those that assume they are not. The first type of
construction algorithm is suitable when an operation is being moved into an existing,
vacant building. The second type of construction algorithm is suitable for “green
field” applications.
 308
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

Even with “green field” applications, however, there is usually a site plan,
which shows the property, surrounding roads, etc. Given the constraints imposed
by the site plan, it is often necessary to construct the new building within a certain
“envelope.” If a construction algorithm of the second type is used, it is often difficult
to ensure that the resulting building will properly fit into such an envelope. With
construction algorithms of the first type, on the other hand, one may model the
above envelope as an “existing building” to obtain a proper fit between the actual
building and the envelope. Primarily for the above reason, the construction algo-
rithms we present in this chapter are of the first type (i.e., they assume the building
dimensions are given).
In the following sections, we describe the overall modeling techniques and/or
methods used in various layout algorithms: namely, the pairwise exchange method,
a graph-based method, CRAFT, BLOCPLAN, MIP, LOGIC, and MULTIPLE. (For ease
of reference, in those cases where the original authors did not use an acronym, we
created our own.) Detailed information such as input data format or output vari-
ables are not presented since such information is relevant and available only if the
reader obtains a copy of the computer program. However, for each algorithm, we
present sufficient detail not only so that the fundamental concepts are covered but
also so that interested readers may develop their own basic implementation of the
algorithm. The algorithms we present are relatively recent ones with a few excep-
tions, such as CRAFT, which we consider because it is one of the first layout algo-
rithms developed and it provides a good platform for the others. Furthermore, all
the algorithms we present require a basic understanding of heuristic search tech-
niques (such as the “steepest-descent” procedure) and the difference between “locally”
and “globally” optimal solutions. We selected these methods based on their concep-
tual contributions to layout planning methods and/or their unique approach to layout
construction/improvement.

6.4.2 Pairwise Exchange Method

In an earlier discussion, we suggested that the majority of layout problems involve
the redesign of an existing layout or existing facility, which is typically triggered by
changes made to the building, kaizen events conducted in lean manufacturing, the
addition of new machines, changes in product mixes, decisions related to the con-
traction and expansion of storage areas, or a simple realization that the old layout is
no longer adequate for the firm’s current needs. Thus, given an existing layout, the
challenge is to develop an improved layout by making feasible or realistic changes
to the existing layout.
The pairwise exchange method ([53], [9]) is an improvement-type layout algo-
rithm. Although it can be used with both an adjacency-based and distance-based
objective, it is often used with the latter. We will illustrate the pairwise exchange
method below through an example based on equal-area departments for simplicity.
Its implementation with unequal-area departments (which is the case in practice)
will be shown later via CRAFT, MULTIPLE, and others.
Consider four departments of equal sizes. The material flow data are given in
Table 6.1. The existing layout is shown in Figure 6.10a. A distance matrix can be
obtained based on the existing layout as given in Table 6.2.
 309
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

Table 6.1 Material Flow Matrix

To Department
1 2 3 4
1 — 10 15 20
From
2 — 10 5
Department
3 — 5
4 —

The distance-based objective function value (or “total cost”) for the existing
layout is computed as follows:
TC1234  10(1)  15(2)  20(3)  10(1)  5(2)  5(1)  125
The subscript notation indicates the order of the departments in the initial
layout.
The pairwise exchange method simply states that for each iteration, all feasi-
ble exchanges in the locations of department pairs are evaluated one at a time, and
the pair that results in the largest reduction in total cost is selected. (Moving in the
direction of largest cost reduction is also known as “steepest descent” in optimiza-
tion.) Evaluating each exchange one at a time means that when departments i and j
are exchanged and the layout cost is computed, departments i and j are placed back
in their current positions in the layout before the next pair of departments is ex-
changed.
Since all the departments are of equal size, and none of the departments has
been declared fixed, the feasible exchanges for the above problem are 1-2, 1-3, 1-4,
2-3, 2-4, and 3-4. The distance matrix is recomputed each time an exchange is per-
formed. The layout costs resulting from the above exchanges are
TC2134(1-2)  10(1)  15(1)  20(2)  10(2)  5(3)  5(1)  105
TC3214(1-3)  10(1)  15(2)  20(1)  10(1)  5(2)  5(3)  95
TC4231(1-4)  10(2)  15(1)  20(3)  10(1)  5(1)  5(2)  120
TC1324(2-3)  10(2)  15(1)  20(3)  10(1)  5(1)  5(2)  120
TC1432(2-4)  10(3)  15(2)  20(1)  10(1)  5(2)  5(1)  105
TC1243(3-4)  10(1)  15(3)  20(2)  10(2)  5(1)  5(1)  125
Using steepest descent, we select pair 1-3 and exchange the locations of de-
partments 1 and 3 in the layout. The new layout is shown in Figure 6.10b.

(a) Iteration 0 1 2 3 4

(b) Iteration 1 3 2 1 4

(c) Iteration 2 2 3 1 4

Figure 6.10 Layouts corresponding to each iteration.

 310
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

Table 6.2 Distance Matrix Based on Existing Layout

To Department
1 2 3 4
1 — 1 2 3
From
Department 2 — 1 2
3 — 1
4 —

For the next iteration, we consider all feasible exchanges that consist of the
same set as in iteration 1. The resulting layout costs are
TC3124(1-2)  10(1)  15(1)  20(2)  10(2)  5(1)  5(3)  105
TC1234(1-3)  10(1)  15(2)  20(3)  10(1)  5(2)  5(1)  125
TC3241(1-4)  10(2)  15(3)  20(1)  10(1)  5(1)  5(2)  110
TC2314(2-3)  10(2)  15(1)  20(1)  10(1)  5(3)  5(2)  90
TC3412(2-4)  10(1)  15(2)  20(1)  10(3)  5(2)  5(1)  105
TC4213(3-4)  10(1)  15(1)  20(2)  10(2)  5(1)  5(3)  105
The pair 2-3 is selected with a total layout cost of 90. Figure 6.10c shows the result-
ing layout after two iterations. Continuing on, the third iteration calculations are
TC1324(1-2)  10(2)  15(1)  20(3)  10(1)  5(3)  5(1)  120
TC2134(1-3)  10(1)  15(1)  20(2)  10(2)  5(3)  5(1)  105
TC2341(1-4)  10(3)  15(2)  20(1)  10(1)  5(2)  5(1)  105
TC3214(2-3)  10(1)  15(2)  20(1)  10(1)  5(2)  5(3)  95
TC4312(2-4)  10(1)  15(1)  20(2)  10(2)  5(3)  5(1)  105
TC2413(3-4)  10(2)  15(1)  20(1)  10(3)  5(1)  5(2)  100
Since the lowest layout cost for the third iteration, 95, is worse than the current lay-
out cost of 90 (obtained in the second iteration), the procedure is terminated. The
final layout arrangement is 2-3-1-4, as shown in Figure 6.10c. The final layout is also
known as a “two-opt layout” since there are no two-way exchanges that can further
reduce the layout cost.
We stress that the pairwise exchange procedure described above is not guar-
anteed to yield the optimal layout because the final outcome is dependent on the
initial layout; that is, a different initial layout often results in a different solution.
Thus, we can only claim local optimality; that is, a two-opt layout is only locally op-
timal; it may or may not be globally optimal. Also, you may have observed that it is
possible to cycle back to one of the alternative layout arrangements from a previous
iteration. For instance, the layout arrangement 1-2-3-4 is what we started with, and
we see the same arrangement in the second iteration when departments 1 and 3
were exchanged based on the solution obtained from iteration 1—that is, layout
arrangement 3-2-1-4. Additionally, symmetric layout arrangements may also occur;
for example, 4-3-2-1 in iteration 3 is identical to 1-2-3-4.
Note that a pairwise exchange can be easily accomplished if the pair of
departments considered are of equal size (as we assumed in this example). Otherwise,
 311
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

we would have to expend some effort in rearranging the two departments being
exchanged and possibly the other departments in the layout. We will discuss the
exchange of unequal-area departments in CRAFT and subsequent algorithms.

6.4.3 Graph-Based Method

The graph-based method is a construction-type layout algorithm; it has its roots in
graph theory. It is often used with an adjacency-based objective. The recognition of
the usefulness of graph theory as a mathematical tool in the solution of facilities
planning problems dates back to the late 1960s [35] and early 1970s [58]. Graph the-
ory methods have similarities with the SLP method developed by Muther [49].
Consider the block layout shown in Figure 6.11a. We first construct an adja-
cency graph, where each node represents a department, with a connecting arc
between two nodes indicating that two departments share a common border. A simi-
lar graph is constructed for the alternative block layout shown in Figure 6.11b. We ob-
serve that the two graphs shown in Figure 6.11 are subgraphs of the graph shown in

0 Arc Weight
9
5 1-2 9
7
1-5 0
2 2-3 12
2 2-4 13
13
2-5 7
12 3-4 20
4 4-5 2
63 (total)
20
3

(a)

5
0
Arc Weight
1
7 1-5 0
9 2-5 7
10 1-2 9
2
1-4 10
13 2-4 13
12 2-3 12
4 3-4 20
71 (total)
20
3

(b)
Figure 6.11 Adjacency graphs for alternative block layouts.
 312
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

φ
1 9
9 1 2
2 8 12 7
12 10
3 13 0 8 13 5
20 7 10
4 0
20 2
2 3 4
5
φ

(a) Relationship chart (b) Relationship diagram

Figure 6.12 Relationship chart and relationship diagram for graph-based example.

Figure 6.12b, which is derived from the relationship chart in Figure 6.12a. The rela-
tionship chart displays numerical “weights” rather than alphabetic closeness ratings.
Given the adjacency-based objective, block layout (b) is better than block
layout (a) with scores of 71 and 63, respectively. Thus, finding a maximally weighted
block layout is equivalent to obtaining an adjacency graph with the maximum sum
of arc weights.
Before we describe a method for determining adjacency graphs, we first make
the following observations:
a. The adjacency score does not account for distance, nor does it account for re-
lationships other than those between adjacent departments.
b. Dimensional specifications of departments are not considered; the length of
common boundaries between adjacent departments is also not considered.
c. The arcs do not intersect; this property of graphs is called planarity. We note that
the graph obtained from the relationship diagram is usually a nonplanar graph.
d. The score is very sensitive to the assignment of numerical weights in the rela-
tionship chart.
There are two strategies we can follow in developing a maximally weighted planar
adjacency graph. One way is to start with the graph from the relationship diagram
and selectively prune connecting arcs while making sure that the final graph is planar.
A second approach is to iteratively construct an adjacency graph via a node insertion
algorithm while retaining planarity at all times. A heuristic procedure based on the
second approach is described below.
6.4.3.1 Procedure
Step 1. From the relationship chart in Figure 6.12a, select the department pair
with the largest weight. Ties, if any, are broken arbitrarily. Thus, depart-
ments 3 and 4 are selected to enter the graph.
Step 2. Next, select the third department to enter. The third department is selected
based on the sum of the weights with respect to departments 3 and 4. From
Figure 6.13a, department 2 is chosen with a value of 25. The columns in this
figure correspond to the departments already in the adjacency graph, and
the rows correspond to departments not yet selected. The last column gives
the sum of the weights for each unassigned department.
 313
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

a. Step 2 2
3 4 Total
1 8 10 18 12 13
2 12 13 25 (best)
5 0 2 2
3 4
20

b. Step 3 2
2 3 4 Total
1 9 8 10 27 (best) 9
5 7 0 9 9 12 13
1

8 10
3 4
20

c. Step 4 2
1 2 3 4
5 0 7 0 2 7
9
13
12 0 5
Faces Total
1
1-2-3 7
1-2-4 9 (best) 2
8 10
1-3-4 2
2-3-4 9 (best) 3 4
20

Figure 6.13 Steps of the graph-based procedure.

Step 3. We then pick the fourth department to enter by evaluating the value of
adding one of the unassigned departments represented by a node on a
face of the graph. A face of a graph is a bounded region of a graph. For
instance, a triangular face is the region bounded by arcs 2-3, 3-4, and 4-2
in Figure 6.13a. We will denote this face as 2-3-4. The outside region is
referred to as the external face. For our example, the value of adding
departments 1 and 5 is 27 and 9, respectively. Department 1 is selected
and placed inside the region 2-3-4, as shown in Figure 6.13b.
Step 4. The remaining task is to determine on which face to insert department 5.
For this step, department 5 can be inserted on faces 1-2-3, 1-2-4, 1-3-4,
and 2-3-4. Inserting 5 on faces 1-2-4 and 2-3-4 yields identical values of 9.
We arbitrarily select 1-2-4. The final adjacency graph is given in Figure 6.13c.
This solution is optimal, with a total sum of arc weights equal to 81.
Step 5. Having determined an adjacency graph, the final step is to construct a cor-
responding block layout. A block layout based on the final adjacency
graph is shown in Figure 6.14. The manner by which we constructed
the block layout is analogous to the SLP method. We should note that in
constructing the block layout, the original department shapes had to be
altered significantly in order to satisfy the requirements of the adjacency
graph. In practice, we may not have as much latitude in making such alter-
ations since department shapes are generally derived from the geometry
of the individual machines within the department and the internal layout
 314
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

2
12
13
Arc Weight
7
9 1-2 9
1-3 8
1-4 10
1 5 1-5 0
0 2-3 12
8 2-4 13
2-5 7
10 2 3-4 20
3 4-5 2
20 81 (total)
4

Figure 6.14 Block layout from the final adjacency graph.

configuration. We will discuss department shapes and their control later in

this chapter (see Section 6.5). Finally, we should point out that there are
algorithmic methods for performing this step as demonstrated by Giffin
et al. [20] and Hassan and Hogg [23].

6.4.4 Craft
Introduced in 1963 by Armour, Buffa, and Vollman (see [3], [9]), the Computerized
Relative Allocation of Facilities Technique (CRAFT) is one of the earliest layout algo-
rithms presented in the literature. It uses a from-to chart as input data for the flow. Lay-
out “cost” is measured by the distance-based objective function shown in Equation 6.1.
Departments are not restricted to rectangular shapes, and the layout is represented in
a discrete fashion.
Since CRAFT is an improvement-type layout algorithm, it starts with an initial
layout, which typically represents the actual layout of an existing facility but may
also represent a prospective layout developed by another algorithm. CRAFT begins
by determining the centroids of the departments in the initial layout. It then calcu-
lates the rectilinear distance between pairs of department centroids and stores the
values in a distance matrix. The initial layout cost is determined by multiplying each
entry in the from-to chart with the corresponding entries in the unit cost matrix (i.e.,
the cij values) and the distance matrix.
CRAFT next considers all possible two-way (pairwise) or three-way depart-
ment exchanges and identifies the best exchange, that is, the one that yields the
largest reduction in the layout cost. (No department can be split as a result of a
two-way or three-way exchange.) Once the best exchange is identified, CRAFT
updates the layout according to the best exchange and computes the new depart-
ment centroids as well as the new layout cost to complete the first iteration. The
next iteration starts with CRAFT once again identifying the best exchange by con-
sidering all possible two-way or three-way exchanges in the (updated) layout.
The process continues until no further reduction in layout cost can be obtained.
The final layout obtained in such a manner is also known as a two-opt (three-
opt) layout since no two-way (three-way) exchanges can further reduce the lay-
out cost.
 315
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

Since computers were relatively “slow” in the 1960s, the original implemen-
tation of CRAFT deviated slightly from the above description. When the program
considered exchanging the locations of departments i and j, instead of actually
exchanging the department locations to compute their new centroids and the ac-
tual layout cost, it computed an estimated layout cost simply by temporaily treat-
ing the centroid of department i in the current layout as the centroid of
department j and vice versa; that is, it simply swapped the centroids of depart-
ments i and j.
The error incurred in estimating the layout cost as described above depends
on the relative size of the two departments being exchanged. If the departments dif-
fer in size, then the estimated centroids may deviate significantly from their correct
locations. (Of course, if the departments are equal in area, no error will be in-
curred.) As a result, the actual reduction in the layout cost may be overestimated or
underestimated. Although it does not fully address the above error, once the best
exchange based on the estimated layout cost is identified, CRAFT exchanges the loca-
tions of the departments and computes their new centroids (and the actual layout
cost) before continuing to the next iteration.
A more important refinement we need to make in the above description of
CRAFT is concerned with the exchange procedure. Although at first it may seem
“too detailed,” this refinement is important from a conceptual point of view since it
demonstrates an intricate aspect of using computers for layout purposes. When
CRAFT considers exchanging two departments, instead of examining all possible
exchanges as we stated above, it actually considers exchanging only those depart-
ments that are either adjacent (i.e., that share a border) or that are equal in area.
Such a restriction is not arbitrary. Given that departments cannot be split, it would
be impossible to exchange two departments without “shifting” the location of the
other departments in the layout, unless the two departments are either adjacent or
equal in area. (Why?) Since CRAFT is not capable of “automatically” shifting depart-
ments in such a manner, it considers exchanging only those that are adjacent or
equal in area.
Obviously, two departments with equal areas, whether they are adjacent or
not, can always be exchanged without shifting the other departments in the layout.
However, if two departments are not equal in area, then adjacency is a necessary
but not sufficient condition for being able to exchange them without shifting the
other departments. That is, in certain cases, even if two (unequal-area) departments
are adjacent, it may not be possible to exchange them without shifting the other de-
partments. We will later present an example for such a case.
We also need to stress that, while searching for a better solution, CRAFT
picks only the best (estimated) exchange at each iteration, which makes it a
“steepest descent” procedure. It also does not “look back” or “look forward” dur-
ing the above search. Therefore, CRAFT will terminate at the first two-opt or
three-opt solution that it encounters during the search. Such a solution is very
likely to be only locally optimal. Furthermore, with such a search procedure, the
termination point (or the final layout) will be strongly influenced by the starting
point (or the initial layout). Consequently, CRAFT is a highly “path-dependent”
heuristic and to use it effectively, we generally recommend trying different initial
solutions (if possible) or trying different exchange options (two-way vs. three-way)
at each iteration.
 316
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

CRAFT is generally flexible with respect to department shapes. As long as

the department is not split, it can accommodate virtually any department shape.
Theoretically, due to the centroid-to-centroid distance measure, the optimum lay-
out (which has an objective function value of zero) consists of concentric rectan-
gles! Of course, the above problem stems from the fact that the centroids of some
O-shaped, U-shaped, and L-shaped departments may lie outside the department
itself. Unless the initial layout contains concentric departments, CRAFT will not
construct such a layout. However, some department shapes may be irregular, and
the objective function value may be underestimated due to the centroid-to-centroid
measure.
CRAFT is normally restricted to rectangular buildings. However, through
“dummy” departments, it can be used with nonrectangular buildings as well.
Dummy departments have no flows or interaction with other departments but re-
quire a certain area specified by the layout planner. In general, dummy departments
may be used to

1. Fill building irregularities.

2. Represent obstacles or unusable areas in the facility (such as stairways, eleva-
tors, plant services, and so on).
3. Represent extra space in the facility.
4. Aid in evaluating aisle locations in the final layout.

Note that, when a dummy department is used to represent an obstacle, its location
must be fixed. Fortunately, CRAFT allows the user to fix the location of any depart-
ment (dummy or otherwise). Such a feature is especially helpful in modeling obsta-
cles, as well as in modeling departments such as receiving and shipping in an
existing facility.
One of CRAFT’s strengths is that it can capture the initial layout with rea-
sonable accuracy. This strength stems primarily from CRAFT’s ability to accom-
modate nonrectangular departments or obstacles located anywhere in a possibly
nonrectangular building. However, in addition to being highly path dependent,
one of CRAFT’s weaknesses is that it will rarely generate department shapes that
result in straight, uninterrupted aisles as is desired in the final layout. Fixing some
departments to specific locations, and in some cases placing dummy departments
in the layout to represent main aisles, may lead to more reasonable department
shapes. Nevertheless, as is the case with virtually all computerized layout algo-
rithms, the final layout generated by the computer should not be presented to the
decision maker before the layout planner “molds” or “massages” it into a practical
layout.
As we demonstrate in the following example, molding or massaging a layout
involves relatively minor adjustments being made to the department shapes and/or
department areas in the final layout. The fact that such adjustments are almost al-
ways necessary does not imply that computerized layout algorithms are of limited
use. To the contrary, by considering a large number of alternatives in a very short
time, a computerized layout algorithm narrows down the solution space for the lay-
out planner, who can then concentrate on further evaluating and massaging a few
“promising” solutions identified by the computer.
 317
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

Example 6.1
Using CRAFT to improve an existing layout
Consider a manufacturing facility with seven departments. The department names, their
areas, and the from-to chart are shown in Table 6.3. We assume that all the cij values are set
equal to 1. The building and the current layout (which we supply as the initial layout to
CRAFT) are shown in Figure 6.15, where each grid is assumed to measure 20  20. Since
the total available space (72,000 ft2) exceeds the total required space (70,000 ft2), we gener-
ate a single dummy department (H) with an area of 2000 ft2. (In most cases, depending on
the amount of excess space available, it is highly desirable to use two or more dummy de-
partments with more or less evenly distributed space requirements. Note that the allocation
of excess space in the facility plays a significant role in determining future expansion options
for many departments.) For practical reasons, we assume that the locations of the receiving
(A) and shipping (G) departments are fixed.
CRAFT first computes the centroids of the departments, which are shown in Figure 6.15.
For each department pair, it then computes the rectilinear distance between their centroids
and multiplies it by the corresponding entry in the from-to chart. For example, the rectilin-
ear distance between the centroids of departments A and B is equal to six grids. CRAFT
multiplies 6 by 45 and adds the result to the objective function. Repeating the above calcu-
lation for all department pairs with nonzero flow yields an initial layout cost of 2974 units.
(We caution the reader that CRAFT computes the distances in grids, not in feet. Therefore,
the actual layout cost is equal to 2974  20  59,480 units.)
Subsequently, CRAFT performs the first iteration and exchanges departments E and F
to obtain the layout shown in Figure 6.16. Departments E and F are not equal in area; how-
ever, since they are adjacent, one can draw a single “box” around E and F such that it con-
tains both departments E and F but no other departments. (Note that if E and F were not
adjacent, drawing such a box would not be possible.) Since the above box contains no
other departments, CRAFT will exchange departments E and F without shifting any other
department. (Naturally, CRAFT does not “draw boxes.” We introduced the box analogy only
to clarify our description of CRAFT.)
There are several ways in which departments E and F may be exchanged within the
above box. Comparing the locations of departments E and F in Figures 6.15 and 6.16, it is
evident that CRAFT started with the leftmost column of department F (the larger of the two
departments) and labeled the first 20 grids of department F as department E. To complete
the exchange, all the grids originally labeled with an E have been converted to an F. Of
course, one may implement the above scheme starting from the rightmost column (or, say,
the top row) of department F. Regardless of its exact implementation, however, such an

Table 6.3 Departmental Data and From-To Chart for Example 6.1
Department Area No. of Flow
Name (ft2) Grids A B C D E F G H
1. A: Receiving 12,000 30 0 45 15 25 10 5 0 0
2. B: Milling 8000 20 0 0 0 30 25 15 0 0
3. C: Press 6000 15 0 0 0 0 5 10 0 0
4. D: Screw m/c 12,000 30 0 20 0 0 35 0 0 0
5. E: Assembly 8000 20 0 0 0 0 0 65 35 0
6. F: Plating 12,000 30 0 5 0 0 25 0 65 0
7. G: Shipping 12,000 30 0 0 0 0 0 0 0 0
8. H: Dummy 2000 5 0 0 0 0 0 0 0 0
 318
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
1 A A A A A A A A A A G G G G G G G G
2 A A G G
3 A A A A A A A A A A G G G G
4 B B B B B C C C C C E E G G G G G G
5 B B C C E E E E E E E E
6 B B C C C C C E E E E E E E E
7 B B B B B D D D D F F F F F F F E E
8 D D D D D D D F F F F
9 D D D F F F F F F
10 D D D D D D D D H H H H H F F F F F

Figure 6.15 Initial CRAFT layout and department centroids for

Example 6.1 (z  2974  20  59,480 units).

exchange scheme generally leads to poor department shapes. In fact, the department
shapes in CRAFT often have a tendency to deteriorate with the number of iterations even if
all the departments in the initial layout are rectangular.
The estimated reduction in the layout cost obtained by exchanging (the centroids of )
departments E and F is equal to 202 units. Upon exchanging departments E and F and com-
puting the new department centroids, CRAFT computes the actual cost of the layout shown
in Figure 6.16 as 2953 units. Hence, the actual reduction in the layout cost is 21 units as op-
posed to 202 units. The reader may verify that the above significant deviation is largely due
to the fact that the new centroid of department F (after the exchange) deviates substantially
from its estimated location.
In the next iteration, based on an estimated reduction of 95 units in the layout cost,
CRAFT exchanges departments B and C to obtain the layout shown in Figure 6.17. The layout
cost is equal to 2833.50 units, which represents an actual reduction of 119.50 units. This clearly
illustrates that the error in estimation may be in either direction. Using estimated costs, CRAFT
determines that no other (equal-area or adjacent) two-way or three-way exchange can further
reduce the cost of the layout and it terminates with the final solution shown in Figure 6.17.

A computer-generated layout should not be presented to the decision maker

before the analyst molds or massages it into a practical layout. In massaging a layout,
the analyst may generally disregard the grids and use a continuous representation. In
so doing, he or she may smooth the department borders and slightly change their

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
1 A A A A A A A A A A G G G G G G G G
2 A A G G
3 A A A A A A A A A A G G G G
4 B B B B B C C C C C F F G G G G G G
5 B B C C F F F F F F F F
6 B B C C C C C F F F F F F F
7 B B B B B D D D D E E E E E E F F
8 D D D D D D D E E F F
9 D D D E E E E E E F F
10 D D D D D D D D H H H H H E E F F F

Figure 6.16 Intermediate CRAFT layout obtained after exchanging departments

E and F (z  2953  20  59,060 units).
 319
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
1 A A A A A A A A A A G G G G G G G G
2 A A G G
3 A A A A A A A A A A G G G G
4 C C C B B B B B B B F F G G G G G G
5 C C C B B F F F F F F F F
6 C C B B B B B B F F F F F F F
7 C C C C B D D D D E E E E E E F F
8 D D D D D D D E E F F
9 D D D E E E E E E F F
10 D D D D D D D D H H H H H E E F F F

Figure 6.17 Final CRAFT layout (z  2833.50  20  56,670 units).

areas or orientations, if necessary. After massaging the layout shown in Figure 6.17,
we obtained the layout shown in Figure 6.18. Note that, other than to department H
(which is a dummy department), we made no significant changes to any of the de-
partments; yet the layout shown in Figure 6.18 is more reasonable and perhaps more
practical than the one shown in Figure 6.17.
Once it is massaged in the above manner, a layout cannot be generally reeval-
uated via a computer-based layout algorithm unless one is willing to redefine the
grid size and repeat the process using the new grid size. Also, we need to stress that
in many real-world problems, layout massaging usually goes beyond adjusting
department shapes or areas. In massaging a layout, the analyst must often take
into account certain qualitative factors or constraints that may not have been con-
sidered by the algorithm.
Earlier we remarked that if two departments are not equal in area, then adja-
cency is a necessary but not sufficient condition for being able to exchange them
without shifting the other departments. Obviously, adjacency is necessary since it is
otherwise physically impossible to exchange two unequal-area departments without
shifting other departments. (Recall that extra space is also modeled as a “depart-
ment.”) The fact that adjacency is not sufficient, on the other hand, can be shown
through the following example [40].
Consider a 7  5 layout with six departments as shown in Figure 6.19. Note
that departments 2 and 4 are not equal in area but they are adjacent, which implies
that we can draw a single box around them. Yet one cannot exchange departments

A G

C B
F

D H E

Figure 6.18 Final “massaged” layout obtained with CRAFT.

 320
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

6 6 6 5 5
6 6 6 5 5
6 6 6 5 4
6 6 6 4 4
2 2 2 2 2
1 1 2 3 3
1 1 2 3 3

Figure 6.19 Example to show that CRAFT may not be able to exchange two adjacent
departments that are not equal in area.

2 and 4 without splitting department 2. In fact, if we give the above layout to CRAFT
and fix the locations of all the departments except 2 and 4, CRAFT does not ex-
change departments 2 and 4 even if f1,4 is set equal to a large value [40]. The above
example is, of course, carefully constructed to show that adjacency is not sufficient.
In most cases, two unequal-area departments that are adjacent can be exchanged
without splitting either one.
Specifying the conditions required for a three-way exchange is somewhat
more complicated. Suppose departments i, j, and k are considered for a three-way
exchange such that department i “replaces” j, department j “replaces” k, and depart-
ment k “replaces” i. If a single “box” can be drawn around departments i, j, and k,
and this “box” does not contain any other department, then (except for cases such
as the one shown above for departments 2 and 4) we can perform a three-way ex-
change without shifting any of the other departments. Note that, in order to draw
such a box, each of the three departments need not share a border with the other
two; one may still draw the above box if department i is adjacent to department j
(but not k) and department k is adjacent to department j (but not i).
If it is not possible to draw such a box, equal-area departments may still permit
some three-way exchanges. Suppose departments i and j are adjacent but department
k is separated from both. For the above exchange involving departments i, j, and k, for
example, one may perform the exchange without shifting any of the other departments
if departments j and k are equal in area. Of course, other combinations (including the
case where all three departments are nonadjacent but equal in area) are also possible.
Computer implementation of three-way exchanges (i.e., deciding which department to
move first and how to reassign the grids) is not straightforward. Furthermore, the num-
ber of possible three-way exchanges increases quite rapidly with the number of de-
partments (which may result in long execution times). Since more recent (single-floor)
facility layout algorithms focus on two-way exchanges only, and since the adjacency or
equal-area requirement has been relaxed with various layout formation techniques, we
will not present in detail three-way exchanges performed by CRAFT.
A personal computer implementation of CRAFT by Hosni, Whitehouse, and
Atkins [28] was distributed by the Institute of Industrial Engineers (IIE) under the name
MICRO-CRAFT (Since IIE discontinued software distribution, the reader may refer to
the original authors to inquire about the code.) MICRO-CRAFT (or MCRAFT) is similar
to CRAFT except that the above constraint is relaxed (i.e., MCRAFT can exchange any
two departments whether they are adjacent or not). Such an improvement is obtained
by using a layout formation technique that “automatically” shifts other departments
when two unequal-area, nonadjacent departments are exchanged. Instead of assigning
each grid to a particular department, MCRAFT first divides the facility into “bands,” and
 321
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

the grids in each band are then assigned to one or more departments. The number of
bands in the layout is specified by the user. MCRAFT’s layout formation technique—
which was originally used in an earlier algorithm, namely, Automated Layout Design
Program (ALDEP) [57]—is described through the following example.

Example 6.2
Band-based layout formation used in MCRAFT
Consider the same data given for Example 6.1. Unlike CRAFT (where the user selects the
grid size), MCRAFT asks for the length and width of the building and the number of bands.
The program then computes the appropriate grid size and the resulting number of rows and
columns in the layout. Thus, setting the building length (width) equal to 360 feet (200 feet),
and the number of bands equal to, say, three, we obtain the initial layout shown in Figure 6.20,
where each band is six rows wide. MCRAFT forms a layout by starting at the upper-lefthand
corner of the building and “sweeping” the bands in a serpentine fashion. In doing so, it follows
a particular sequence of department numbers, which we will refer to as the layout vector
or the fill sequence. The layout shown in Figure 6.20 was obtained from the layout vector
1-7-5-3-2-4-8-6, which is supplied by the user as the initial layout vector. (Note that MCRAFT
designates the departments by numbers instead of letters; department 1 represents depart-
ment A, department 2 represents department B, etc.)
MCRAFT performs four iterations (i.e., four two-way exchanges) before terminating
with the two-opt layout. The departments exchanged at each iteration and the corresponding
layout cost are given as follows: first iteration—departments C and E (59,611.11 units); second
iteration—departments C and H (58,083.34 units); third iteration—departments C and D
(57,483.34 units); and fourth iteration—departments B and C (57,333.34 units). The resulting
three-band layout is shown in Figure 6.21, for which the layout vector is given by 1-7-8-5-3-2-
4-6. Except for department 2, the departments in Figure 6.21 appear to have reasonable
shapes. Unlike CRAFT, the department shapes obtained from the sweep method tend to be
reasonable if an appropriate number of bands is selected. Of course, alternative initial and fi-
nal layouts may be generated by varying the number of bands and the initial layout vector.

Due to the above layout formation technique, MCRAFT cannot capture the ini-
tial layout accurately unless the departments are already arranged in bands. As a
result, one may have to massage the initial layout to make it compatible with

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 1 1 1 1 1 1 1 1 7 777777777777777777777777777777777
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 1 1 1 1 1 1 1 1 7 777777777777777777777777777777777
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 1 1 1 1 1 1 1 1 7 777777777777777777777777777777777
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 1 1 1 1 1 1 1 1 7 777777777777777777777777777777777
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 1 1 1 1 1 1 1 1 7 777777777777777777777777777777777
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 1 1 1 1 1 1 1 1 7 777777777777777777777777777777777
4 4 4 4 4 2 2 2 2 22 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3333333333 55555555555555555555555
4 4 4 4 4 2 2 2 2 22 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3333333333 55555555555555555555555
4 4 4 4 4 2 2 2 2 22 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3333333333 55555555555555555555555
4 4 4 4 4 2 2 2 2 22 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3333333333 55555555555555555555555
4 4 4 4 4 2 2 2 2 22 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3333333333 55555555555555555555555
4 4 4 4 4 2 2 2 2 22 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3333333333 55555555555555555555555
4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 8 8 8 8 8 8 666666666666666666666666666666666
4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 8 8 8 8 8 8 666666666666666666666666666666666
4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 8 8 8 8 8 8 666666666666666666666666666666666
4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 8 8 8 8 8 8 666666666666666666666666666666666
4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 8 8 8 8 8 8 666666666666666666666666666666666
4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 8 8 8 8 8 8 666666666666666666666666666666666

Figure 6.20 Initial MCRAFT layout for Example 6.2 (z  60,661.11 units).
 322
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

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 1 1 1 1 1 1 1 1 7 777777777777777777777777777777777
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 1 1 1 1 1 1 1 1 7 777777777777777777777777777777777
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 1 1 1 1 1 1 1 1 7 777777777777777777777777777777777
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 1 1 1 1 1 1 1 1 7 777777777777777777777777777777777
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 1 1 1 1 1 1 1 1 7 777777777777777777777777777777777
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 1 1 1 1 1 1 1 1 7 777777777777777777777777777777777
2 2 2 2 2 2 2 2 2 22 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 3 3 3 3 3333555555 55555555555555555888888
2 2 2 2 2 2 2 2 2 22 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 3 3 3 3 3333555555 55555555555555555888888
2 2 2 2 2 2 2 2 2 22 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 3 3 3 3 3333555555 55555555555555555888888
2 2 2 2 2 2 2 2 2 22 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 3 3 3 3 3333555555 55555555555555555888888
2 2 2 2 2 2 2 2 2 22 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 3 3 3 3 3333555555 55555555555555555888888
2 2 2 2 2 2 2 2 2 22 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 3 3 3 3 3333555555 55555555555555555888888
2 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 666666666666666666666666666666666
2 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 666666666666666666666666666666666
2 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 666666666666666666666666666666666
2 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 666666666666666666666666666666666
2 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 666666666666666666666666666666666
2 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 666666666666666666666666666666666

Figure 6.21 Final MCRAFT layout (z  57,333.34 units).

MCRAFT. (The reader may compare the original initial layout shown in Figure 6.15
with MCRAFT’s initial layout shown in Figure 6.20.) The actual cost of the initial
layout as computed by CRAFT is equal to 2974  20  59,480 units, whereas
MCRAFT’s initial layout cost is equal to 60,661.11 units—a relatively small difference
considering the fact that the “actual” layout used by CRAFT is itself an approxima-
tion of reality. Based on the number of departments and the actual initial layout,
however, the above difference may be significant for some problems.
Another limitation of the sweep method used in MCRAFT is that the band
width is assumed to be the same for all the bands. As a result, MCRAFT is generally
not as effective as CRAFT in treating obstacles and fixed departments. If there is an
obstacle in the building, the analyst must ensure that its width does not exceed the
band width. Otherwise, the obstacle must be divided into two or more pieces,
which is likely to further complicate matters. In addition, while it is straightforward
in MCRAFT to fix the location of any department, a fixed department may still
“shift” or “float” when certain non-equal-area departments are exchanged.
In the above example, the two fixed departments (1 and 7) remained at their cur-
rent locations since they are the first two departments in the initial layout vector 1-7-5-
3-2-4-8-6. However, if we fix, say, department 2 in Figure 6.20, and the algorithm
exchanges departments 3 and 4, then the location of department 2 will shift when the
layout is formed with the new layout vector. (Why?) In general, when two unequal-area
departments are exchanged, the location of a fixed department will shift if the two de-
partments fall on either side of the fixed department in the layout vector. Considering
that obstacles are also modeled as fixed departments (with zero flow), the above limi-
tation implies that obstacles may shift as well. It is instructive to note that MCRAFT’s pri-
mary strength (i.e., being able to “automatically” shift other departments as necessary)
is also its primary weakness (i.e., fixed departments and obstacles may also be shifted).

6.4.5 BLOCPLAN
BLOCPLAN, which was developed by Donaghey and Pire ([12], [51]), is similar to
MCRAFT in that departments are arranged in bands. However, there are certain
differences between the two algorithms. BLOCPLAN uses a relationship chart as
 323
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

well as a from-to chart as input data for the “flow.” Layout “cost” can be measured
either by the distance-based objective (see Equation 6.1) or the adjacency-based
objective (see Equation 6.2). Furthermore, in BLOCPLAN, the number of bands is
determined by the program and limited to two or three bands. However, the band
widths are allowed to vary. Also, in BLOCPLAN, since each department occupies
exactly one band, all the departments are rectangular in shape. Lastly, unlike
MCRAFT, BLOCPLAN uses the continuous representation.
BLOCPLAN may be used both as a construction algorithm and as an improve-
ment algorithm. In the latter case, as with MCRAFT, it may not be possible to cap-
ture the initial layout accurately. Nevertheless, improvements in the layout are
sought through (two-way) department exchanges. Although the program accepts
both a relationship chart and a from-to chart as input, the two charts can be used
only one at a time when evaluating a layout. That is, a layout is not evaluated ac-
cording to some “combination” of the two charts.
BLOCPLAN first assigns each department to one of the two (or three) bands.
Given all the departments assigned to a particular band, BLOCPLAN computes the
appropriate band width by dividing the total area of the departments in that band
by the building length. The complete layout is formed by computing the appropri-
ate width for each band as described above and arranging the departments in each
band according to a particular sequence.
The above layout formation procedure and the layout scores computed by
BLOCPLAN are explained through the following example.

Example 6.3
Using BLOCPLAN to improve an existing layout
Consider the same data given for Example 6.1. As in MCRAFT, BLOCPLAN will not be able
to capture the initial layout accurately unless the departments are already arranged in
bands. As a result, the analyst may have to massage the initial layout to make it compatible
with BLOCPLAN. (The reader may compare the actual initial layout shown in Figure 6.15
with BLOCPLAN’s initial layout shown in Figure 6.22.) The actual cost of the initial layout as
computed by CRAFT is equal to 59,480 units, whereas BLOCPLAN’s initial layout cost is
equal to 61,061.70 units—a relatively small difference of less than 3%.
BLOCPLAN offers the analyst a variety of options for improving the layout. The ana-
lyst may try some two-way exchanges simply by typing the department indices to be ex-
changed, or he or she may select the “automatic search” option to have the algorithm
generate a prespecified number of layouts. (According to [12], the automatic search option
is based on the “procedures that an experienced BLOCPLAN user used in obtaining a ‘good’
layout”; no details are provided on what these procedures are.) Using BLOCPLAN’s improve-
ment algorithm, which interactively considers all possible two-way exchanges, we obtain
the layout shown in Figure 6.23, where departments C and H have been exchanged. Since
no other two-way exchange leads to a reduction in layout cost, BLOCPLAN was terminated
with the final layout as shown in Figure 6.23.
The “cost” of the final layout, as measured by the distance-based objective and the
from-to chart given for the example problem, is equal to 58,133.34 units. BLOCPLAN implic-
itly assumes that all the cij are equal to 1.0; that is, the user cannot enter a cost matrix. (Re-
call that in the example problem all the cij are assumed to be equal to 1.0.) The final layout
shown in Figure 6.23 can also be evaluated with respect to the adjacency-based objective.
Using Equation 6.2, we compute z  235 units, which is obtained by adding the fij values
 324
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

180' 180'

A G 66.67'

130.91' 98.18' 130.91'

B C E 61.11'

D H F 72.22'

166.15' 27.69' 166.15'

Figure 6.22 Initial BLOCPLAN layout for Example 6.3 (z  61,061.70 units).

between all department pairs that are adjacent in Figure 6.23. The normalized adjacency
score (or the efficiency rating) given by Equation 6.3 is equal to 235/(235  200)  0.54.

If the input data are given as a relationship chart instead of a from-to chart, the
adjacency score may still be computed, provided that a numerical value is assigned
to each closeness rating. More specifically, in place of fij shown in Equation 6.2, one
may use the numerical value of the closeness rating assigned to departments i and
j. The following default values are used in BLOCPLAN: A  10, E  5, I  2, O 
1, U  0, and X  10. The user may of course specify different numerical values.
Although the values A  6, E  5, I  4, O  3, U  2, and X  1 have been used
in some algorithms or texts, for the purposes of computing the adjacency score, using
a nonzero value for U and a nonnegative value for X would not be appropriate.
Department pairs with a U relationship should not affect the score whether they are
adjacent or not, while those pairs with an X relationship should adversely affect the
score if they are adjacent.
Even if a from-to chart is provided, BLOCPLAN computes the adjacency score
based on the relationship chart. This is accomplished by converting the from-to

180' 180'

A G 66.67'

160' 40' 160'

B H E 50'

D C F 83.33'

144' 72' 144'

Figure 6.23 Final BLOCPLAN layout (z  58,133.34 units).
 325
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

chart into a relationship chart and using the above numerical values assigned to the
closeness ratings. Although one may develop alternate schemes for such a conver-
sion, the one used in BLOCPLAN is straightforward; it is explained through the fol-
lowing example.

Example 6.4
Converting a from-to chart into a relationship chart
Consider the from-to chart given earlier for Example 6.1 (see Table 6.3). Since BLOCPLAN
implicitly assumes that all the unit costs are equal to 1.0, and all the distances are symmetric
by definition (i.e., dij  dji for all i and j), it first converts the from-to chart into a flow-
between chart by adding fij to fji for all department pairs. That is, the from-to chart shown
in Table 6.3 is converted into the flow-between chart shown in Table 6.4a. A flow-between
chart is symmetric, by definition.
The maximum flow value in the flow-between chart is equal to 90 units. Dividing the
above maximum by 5 we obtain 18, which leads to the following intervals and correspon-
ding closeness ratings: 73 to 90 units (A), 55 to 72 units (E), 37 to 54 units (I), 19 to 36 units
(O), and 0 to 18 units (U). That is, BLOCPLAN divides the flow values into five intervals of
18 units each. Any flow value between 73 and 90 units is assigned an A relationship, and so
on. Applying this conversion scheme to the above flow-between chart yields the relation-
ship chart shown in Table 6.4b.

Given the default numerical values assigned to the closeness ratings (i.e., A  10,
E  5, I  2, O  1, U  0, X  10) and the relationship chart shown in Table 6.4b,
we compute a normalized adjacency score of 30/48  0.63 for the final layout ob-
tained in Example 6.3 (see Figure 6.23). It is instructive to note that the normalized
adjacency score for the initial layout in Example 6.3 is also equal to 0.63, while the
distance-based objectives for the initial and final layouts in Example 6.3 are equal to
61,061.70 and 58,133.34 units, respectively. The above result illustrates the concern
we expressed earlier for the adjacency score; that is, it is possible for two layouts to
have virtually the same (normalized) adjacency score but different travel distances
for parts flow. BLOCPLAN reports both the normalized adjacency score and the
distance-based objective for each layout it generates.
In addition to the distance-based layout cost (using the flow-between chart)
and the normalized adjacency score (based on the relationship chart), BLOCPLAN
computes a “REL-DIST” score, which is a distance-based layout cost that uses the
numerical closeness ratings instead of the flow values. That is, the numerical values

Table 6.4 Flow-Between Chart and Relationship Chart for Example 6.4
A B C D E F G H A B C D E F G H
A 0 45 15 25 10 5 0 0 A — I U O U U U U
B 0 0 50 25 20 0 0 B — U I O O U U
C 0 0 5 10 0 0 C — U U U U U
D 0 35 0 0 0 D — O U U U
E 0 90 35 0 E — A O U
F 0 65 0 F — E U
G 0 0 G — U
H 0 H —
(a) Flow-between chart (b) Relationship chart
 326
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

of the closeness ratings are multiplied by the rectilinear distances between depart-
ment centroids. For example, for the initial (final) layout shown in Figure 6.22
(Figure 6.23), the REL-DIST score is equal to 2887.01 (2708.33) units. The REL-DIST
score is useful when a from-to chart is not available and the layout must be evalu-
ated with respect to a relationship chart only.
BLOCPLAN also computes a normalized REL-DIST score. However, the up-
per and lower bounds used in normalizing the REL-DIST score depend on the
particular layout being evaluated (we refer the reader to [12]). As a result, caution
must be exercised when comparing the normalized REL-DIST scores of two com-
peting layouts.
We remarked earlier that BLOCPLAN implicitly assumes that all the unit costs
are equal to 1.0. This assumption is not as restrictive as it may first seem. If the unit
costs are not equal to 1.0, then provided that they are symmetric, that is, cij  cji for
all i and j, the user may first multiply each flow value with the corresponding unit
cost and then enter the result as the “flow.” Note that unit costs must be symmetric
since BLOCPLAN works with a flow-between chart as opposed to a from-to chart as
shown earlier in Example 6.4.
BLOCPLAN may also be used as a construction algorithm. In doing so, the lay-
out planner may indicate the location of certain departments a priori. As we re-
marked earlier, whether one inputs the entire initial layout or a few departments
that must be located prior to layout construction, it may not be possible to accu-
rately locate all the departments. BLOCPLAN uses the following scheme to specify
the location of one or more departments: the entire building is divided into nine cells
(labeled A through I) that are arranged in three bands. The top band contains
cells A, B, and C (from left to right), while the middle and bottom bands contain
cells D, E, F and G, H, I (from left to right), respectively. Each cell is further divided
into two halves: a right half and a left half. Hence, there are a total of 18 possible
locations, which is also equal to the maximum number of departments that BLOC-
PLAN will accept.
The location of a department is designated by indicating the appropriate side
of a cell. For example, to locate a department on the northeast corner of the build-
ing, one would use C-R to indicate the right side of cell C. To locate a department,
say, on the west side of the building, one would use D-L, and so on. Of course, re-
gardless of the cell they are assigned to, all departments will be arranged in two or
three bands when BLOCPLAN constructs a layout. Each department may be as-
signed to only one of the 18 cells.

6.4.6 MIP1
Using the continuous representation, the facility layout problem may be formulated
as a mixed integer programming (MIP) problem if all the departments are assumed
to be rectangular. Recall that, with rectangular departments, just the centroid and
the length (or width) of a department fully define its location and shape. Although
an alternate objective may be used, the model we show here assumes a distance-
based objective (given by Equation 6.1) with dij defined as the rectilinear distance

1
The MIP model presented in this section requires a basic knowledge of linear and integer programming.
 327
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

between the centroids of departments i and j. Generally speaking, models based on

mathematical programming are regarded as construction-type layout models since
there is no need to enter an initial layout. However, as we discuss later, such mod-
els may also be used to improve a given layout.
The model we show here is based on the one presented by Montreuil [45]. A
similar model is also presented by Heragu and Kusiak [25], where the department
dimensions are treated as parameters with known values instead of decision vari-
ables. Assuming that all department dimensions are given and fixed is appropriate
in “machine layout” problems, where each “department” represents the rectangular
“footprint” of a machine. Note that, with such an approach, the layout model is ac-
tually used as a two-dimensional “packing” algorithm to determine the locations of
rectangular objects with known shapes. Aside from possible differences in the ob-
jective function, such a packing problem is also known as the “two-dimensional bin
packing” problem [8].
Treating the department dimensions as decision variables, the facility layout
problem may be formulated as follows. Consider first the problem parameters. Let
Bx be the building length (measured along the x-coordinate),
By be the building width (measured along the y-coordinate),
Ai be the area of department i,
Lil be the lower limit on the length of department i,
Lui be the upper limit on the length of department i,
W il be the lower limit on the width of department i,
W ui be the upper limit on the width of department i,
M be a large number.
Consider next the decision variables. Let
i be the x-coordinate of the centroid of department i,
i be the y-coordinate of the centroid of department i,
xi be the x-coordinate of the left (or west) side of department i,
x i be the x-coordinate of the right (or east) side of department i,
yi be the y-coordinate of the top (or north side) of department i,
y i be the y-coordinate of the bottom (or south side) of department i,
z ijx be equal to 1 if department i is strictly to the east of department j, and 0
otherwise,
z ijy be equal to 1 if department i is strictly to the north of department j, and 0
otherwise.
Note that department i would be strictly to the east of department j if and
only if x j xi . Likewise, department i would be strictly to the north of department
j if and only if y j y i. Two departments are guaranteed not to overlap if they are
“separated” along either the x-coordinate (i.e., one of the departments is strictly to
the east of the other) or the y-coordinate (i.e., one of the departments is strictly to
the north of the other). Of course, it is possible that two departments are separated
along both the x- and y-coordinates.
 328
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

The above parameter and variable definitions lead to the following model:
Minimize z = a a fij cij (|i - j| + |i - j|) (6.5)
i j

Subject to Lli … (x i– - x i¿ ) … Lui for all i (6.6)

W il … (y i– - yi¿) … W ui for all i (6.7)
(x i– - xi¿)(y i– - yi¿) = Ai for all i (6.8)
0 … xi¿ … x i– … Bx for all i (6.9)
0 … y¿i … y–i … By for all i (6.10)
i = 0.5x i¿ + 0.5x –i for all i (6.11)
i = 0.5yi¿ + 0.5y –i for all i (6.12)
x j– … xi¿ + M (1 - zijx ) for all i and j, i Z j (6.13)
y j– … yi¿ + M (1 - zijy ) for all i and j, i Z j (6.14)
zijx + z jix + z ijy + z jiy Ú 1 for all i and j, i 6 j (6.15)
i , i Ú 0 for all i (6.16)
xi¿, x i–, yi¿, y i– Ú 0 for all i (6.17)
zijx, zijy 0/1 integer for all i and j, i Z j (6.18)
The objective function given by Equation 6.5 is the distance-based objective shown
earlier as Equation 6.1. Constraint sets 6.6 and 6.7, respectively, ensure that the length
and width of each department are within the specified bounds. The area requirement
of each department is expressed through constraint set 6.8, which comprises the only
nonlinear constraints in the model. Constraint sets 6.9 and 6.10 ensure that the depart-
ment sides are defined properly and that each department is located within the build-
ing in the x and y directions, respectively. Constraint sets 6.11 and 6.12, respectively,
define the x- and y- coordinates of the centroid of each department.
Constraint set 6.13 ensures that x j xi (i.e., that department i is strictly to the
east of department j ) if z ijx  1. Note that, if z ijx  0, constraint set 6.13 is satisfied
whether x j is less than or equal to xi or not. In other words, constraint set 6.13
becomes “active” only when z ijx  1. Constraint set 6.14 serves the same purpose as
constraint set 6.13 but in the y direction. Constraint set 6.15 ensures that no two
departments overlap by forcing a separation at least in the east–west or north–south
direction.2 Lastly, constraint sets 6.16 and 6.17 represent the nonnegativity con-
straints, while constraint set 6.18 designates the binary variables.
Due to constraint set 6.8, the above model is nonlinearly constrained. Further-
more, the objective function contains the absolute value operator. There are alternative
schemes one may employ to approximate constraint 6.8 via a linear function. In [45],
the department area is controlled through its perimeter (which is a linear function of
the length and width of the department) and constraint sets 6.6 and 6.7.
Provided that fij 0 for all i and j, the absolute values in the objective func-
tion may be removed by introducing a “positive part” and a “negative part” for each
term [48]. That is, if we set

2
Actually, given the nature of constraints 6.13 and 6.14, one may rewrite constraint 6.15 as z ijx  z jix 
z ijy  z jiy  1. Such constraints are known as “multiple choice” constraints in integer programming;
that is, only one of the binary variables in the constraint may be set equal to one. Some branch-and-bound
algorithms aim at reducing the computational effort by taking advantage of multiple-choice constraints.
 329
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

i - j = ij+ - ji- and i - j = ij+ - ij- , (6.19)

then |i - j| = ij+ - ij- and |i - j| = ij+ + ij- , (6.20)
where    
ij , ij ,  ij , and  ij are all nonnegative variables.
Hence, the nonlinear model given by Equations 6.5 through 6.18 can be trans-
formed into the following linear MIP problem:

Minimize z = a a fij cij (ij+ + ij- + ij+ + ij- ) (6.21)

i j

Subject to Lli … (x i– - xi¿) … Lui for all i (6.22)

W il … (y i– - yi¿) … W ui for all i (6.23)
Pi¿ … (x i– - xi¿ + y i– - yi¿) … P ui for all i (6.24)
0 … xi¿ … x –i … Bx for all i (6.25)
0 … y¿i … y –i … By for all i (6.26)
i = 0.5xi¿ + 0.5x –i for all i (6.27)
i = 0.5yi¿ + 0.5y i– for all i (6.28)
i - j = ij+ - ij- for all i and j, i Z j (6.29)
i - j = ij+ - ij- for all i and j, i Z j (6.30)
x i– … xi¿ + M (1 - zijx ) for all i and j, i Z j (6.31)
y i– … yi¿ + M(1 - zyij ) for all i and j, i Z j (6.32)
zij + zjix + zijy + zjiy Ú 1
x
for all i and j, i 6 j (6.33)
i, i Ú 0 for all i (6.34)
xi¿, x i–, yi¿, y i– Ú 0 for all i (6.35)
ij , ij- , ij+ , ij- Ú 0
+
for all i and j, i Z j (6.36)
zijx, zijy 0/1 integer for all i and j, i Z j (6.37)
where the P il and P ui that appear in constraint set 6.24 represent the lower and up-
per limits imposed on the perimeter of department i, respectively.
In the above formulation, department shapes as well as their areas are indirectly
controlled through constraint sets 6.22 through 6.24. In some cases, one may wish to
explicitly control the shape of a department by placing an upper limit on the ratio of
its longer side to its shorter side. If we designate the upper limit by Ri ( 1.0), such a
ratio may be controlled simply by adding the following (linear) constraints to the
model for department i:

(x i– - xi¿) … Ri (y i– - yi¿) (6.38)

(y i– - yi¿) … Ri (x i– - xi¿) (6.39)

Note that, depending on which side is the longer one, at most one of the above two
constraints will be “binding” or hold as an equality for nonsquare departments.
We remarked earlier that the layout planner may represent an X relationship
between departments i and j by assigning a negative value to fij. However, the
transformation we used earlier to linearize the model (see Equation 6.19) requires
that fij 0 for all i and j. While this requirement rules out the use of “negative
flow” values, it does not imply that the above model cannot effectively capture an
X relationship between two departments. One may insert a minimum required
 330
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

distance between two departments i and j by modifying constraints 6.31 and 6.32
as follows:
x –j + ¢ xij … x i¿ + M(1 - zxij ) for all i and j, i Z j (6.40)
y –j + ¢ yij … yi ¿ + M (1 - zxij ) for all i and j, i Z j (6.41)
where ijx and ijy denote the minimum required distance (or clearance) between de-
partments i and j along the x- and y-coordinates, respectively. Such clearances should
be used sparingly; otherwise, the number of feasible solutions can be very limited.
Also, note that if departments i and j are separated along, say, only the x-coordinate,
then the resulting clearance between them along the y-coordinate may be less than ijy.
Hence, provided that the departmental area requirements need not be satisfied
with precision, the optimum layout with rectangular departments can be, in theory,
obtained by solving the MIP model given by Equations 6.21 through 6.37. For real-
world problems, obtaining an exact solution to the above model is not straightfor-
ward due to the large number of binary variables involved. However, as further
research on MIP and similar mathematical programming-based layout models is con-
ducted, both the problem size (expressed as the number of departments) that can be
tackled and the accuracy with which the department area constraints can be satisfied
have been improving. Of course, faster computers and better optimization packages
are also playing a role. The interested reader may refer to [43], [59], and [1] for details.
To use the MIP model in practice (where the number of departments can eas-
ily exceed 18), one may aim for a heuristic solution rather than the optimal solution.
One obvious heuristic approach is to terminate the branch-and-bound search when
the difference between the least lower bound and the incumbent solution is less
than, say, 5 or 10%. The advantage of such an approach is that the resulting solution
is guaranteed to be within 5 or 10% of the optimum; the primary drawback is that it
may still take a substantial amount of computer time to identify such a solution.
Another possible heuristic approach [46] is to determine the north–south and
east–west relationships between the departments a priori by constructing a “design
skeleton.” With such an approach, first the zijx and the zijy values are determined heuris-
tically, and the MIP model shown by Equations 6.21 through 6.37 can then be solved
as a linear programming problem to “pack” the departments within the building.

Example 6.5
Constructing an optimum layout using the MIP model
Consider the same data given for Example 6.1. Assuming that departments A and G are
fixed on the north side of the building (see the initial layout shown in Figure 6.15), we use
the MIP model to construct the optimal layout. The solution is shown in Figure 6.24. The
layout cost is equal to 53,501.17 units. Although some department areas in Figure 6.24 are
not exactly equal to the values specified in Table 6.3, the reader may verify that the maxi-
mum deviation is only 0.25% (due to department D). Of course, some shape constraints
were imposed on the departments to avoid long-and-narrow departments.

There are many real-world applications where the ideal shape of a depart-
ment is a rectangle. However, there are also cases where L-shaped or U-shaped de-
partments may be more appropriate or necessary. Furthermore, if there are fixed
departments or obstacles in the building, enforcing rectangular department shapes
with exact area values may make it impossible to find a feasible solution (even if
 331
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

180' 180'

66.67' A G

B F 63.43'

125.94' 189.06'
133.33' C

D E H 69.90'

45' 171.25' 114.46' 28.61'

0.68'
Figure 6.24 The optimal layout obtained from the MIP model (z *  53,501.17 units).

the continuous representation is used and the fixed departments and obstacles are
rectangular).
For example, consider a 10  10 building and four departments with the fol-
lowing area requirements: A1  18, A2  20, A3  27, and A4  33. Note that the
total area requirement is actually less than 100 units. If none of the departments
are fixed, then the above problem obviously has many feasible solutions. How-
ever, if the first department is fixed, say, at the northeast corner of the building
(see Figure 6.25a), then there is no feasible solution to the problem. In contrast, if
the area of, say, department 3 or 4 can be slightly adjusted, then three feasible
solutions can be constructed as shown in Figure 6.25a, b, and c.
Thus, retaining some flexibility in departmental area requirements is necessary for
the effective use of the MIP model. In fact, approximating the departmental areas
through some linear relaxation of constraint 6.8 serves a dual purpose: it allows one to
remove a nonlinear constraint, and at the same time it significantly reduces the likeli-
hood of having no feasible solutions when obstacles and/or fixed departments are pres-
ent. Of course, generally speaking, it also increases the number of feasible solutions.
The MIP model may also be used to improve a given layout after specifying
the dimensions and location of fixed departments in the initial layout. This is accom-
plished simply by setting i, i, xi, x i , yi, and y i equal to their appropriate values for
all the fixed departments and/or obstacles. (Nonrectangular obstacles may be repre-
sented as a collection of rectangular dummy departments.) Note that, with such an
approach, the initial layout is “improved” not by exchanging department locations
but rather by determining the new locations of all the (nonfixed) departments at
once. Since such an approach does not explicitly consider “incremental” improvements
to the layout and the resulting relocation costs versus material handling savings, it
may not be a practical approach.
If relocation costs are significant, the layout planner may systematically fix and
unfix certain subsets of the (nonfixed) departments and solve the MIP model several
times to affect incremental changes in the layout. Alternately, the layout planner may
first determine what the north–south/east–west relationships are going to be for two
departments after their exchange, and then subsequently use these relationships and
the above MIP model to “repack” the departments with two of them exchanged.
 332
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

6' 4' 6'

18 3' 18 3'

6.75' 27

20 3.33'
32.86

20 27 7'

3.25' 32.5

3.28' 2.86' 3.86'

(a) (b)

4' 6'

18 3'
5' 20

33 7'

5' 26.43

5.29' 4.71'

(c)

Figure 6.25 Alternative arrangements with relaxed area requirements.

6.4.7 LOGIC
Layout Optimization with Guillotine Induced Cuts (LOGIC) was developed by Tam
[62]. In describing LOGIC, we assume that a from-to chart is given as input data for
the flow. We also assume that layout “cost” is measured by the distance-based ob-
jective function shown in Equation 6.1. The departments generated by LOGIC are
rectangular, provided that the building is rectangular. The layout is represented in a
continuous fashion.
Although LOGIC can be used as a layout improvement algorithm, we will first
present it as a construction algorithm. LOGIC is based on dividing the building into
smaller and smaller portions by executing successive “guillotine cuts”—that is, straight
lines that run from one end of the building to the other. Each cut is either a vertical
cut or a horizontal cut. If a cut is vertical, a department is assigned either to the east
side of the cut or to the west. Given the building width (or a portion of it) and the to-
tal area of all the departments assigned, say, to the east side of a vertical cut, one may
compute the exact x-coordinate of the cut. Likewise, given the building length (or a
portion of it) and the total area of all the departments assigned, say, to the north side
of a horizontal cut, one may compute the exact y-coordinate of the cut.
 333
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

LOGIC executes a series of horizontal and vertical cuts. With each cut, an ap-
propriate subset of the departments are assigned to the east–west or north–south
side of the cut. In order to systematically execute the cuts and keep track of the de-
partments, LOGIC constructs a cut-tree as described in the following example,
where we assume that the cuts and the department assignments are made randomly
for illustration purposes.

Example 6.6
Using LOGIC and cut-trees to construct a layout
Consider the same data given for Example 6.1, except assume none of the departments (in-
cluding A and G) are fixed. That is, assume a vacant building, which measures 360  200
in length and width, respectively. Suppose the first cut is a vertical cut, and departments D, F,
and G are assigned to its east, while the remaining departments are assigned to its west (see
Figure 6.26a). Since the total area required by departments D, F, and G is equal to 36,000 ft2
and the building width is equal to 200 feet, the above vertical cut divides the building into

200' A, B, C, E, H D, F, G

180' 180'
(a)

G 66.67'

111.11' A, B

D, F 133.34'

88.89' C, E, H

180' 180'

(b)

Figure 6.26 Layout obtained by horizontal and vertical cuts executed by LOGIC.
 334
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

66.67' A G 66.67'

44.44' B
D 66.67'

88.89' C, H E
F 66.67'

90' 90' 180'

(c)

66.67' A G 66.67'

44.44' B
D 66.67'

66.67' C
E
F 66.67'

22.22' H
90' 90' 180'

(d)

Figure 6.26 (continued )

two pieces that are 36,000/200  180 feet long each. The first cut is also shown in Figure
6.27, where each node is labeled with a v for a vertical cut and a h for a horizontal cut.
LOGIC next treats each portion of the building as a “building” by itself and repeats
the above procedure until each “building” contains only one department. Consider first
the “building” that contains departments A, B, C, E, and H. Suppose the next cut is a
horizontal cut and that departments A and B are assigned to the north of the cut, while
C, E, and H are assigned to its south. Since the total area required by departments A and
B is equal to 20,000 ft2 and the “building” length is equal to 180 feet, the width of the
“building” which contains departments A and B is equal to 20,000/180  111.11 feet (see
Figure 6.26b).
Consider next the “building” that contains departments D, F, and G. Suppose the third
cut is again a horizontal cut and that department G is assigned to the north of the cut, while
D and F are assigned to its south. Since department G requires an area of 12,000 ft2 and the
“building” length is equal to 180 feet, the width of department G is set equal to 12,000/180 
66.67 feet as shown in Figure 6.26b. The second and third cuts as described above are also
shown in Figure 6.27.
 335
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

A, B, C, D, E, F, G, H

1. v
W E

A, B, C, E, H D, F, G

2. 3.
h h
N S N S

A, B C, E, H G D, F

4. 5. v 6.
h h
N S W E N S

A B C, H E D F

7.
h
N S

C H

Figure 6.27 Cut-tree for Example 6.6.

Suppose the fourth cut horizontally divides departments A and B, while the fifth
cut vertically divides departments {C, H} and E. Further suppose the sixth cut horizontally
divides departments D and F. The layout that results from the above cuts is shown in
Figure 6.26c. Assuming that the seventh and final cut horizontally divides departments C
and H, we obtain the final layout shown in Figure 6.26d. (The above cuts are also shown
in Figure 6.27.)

LOGIC may also be used as an improvement algorithm in a variety of ways.

Here we will show how it can be used to exchange two departments given that
the cut-tree remains the same. Consider the layout shown in Figure 6.26d. Sup-
pose we would like to exchange departments D and E, which are not equal in
area. If the cut-tree remains the same (see Figure 6.27), we simply replace all the
Ds in the tree with Es and vice versa, and compute the new x - and y -coordinates
of the cuts. The resulting layout is shown in Figure 6.28.
It is instructive to note that, since departments D and E are adjacent, CRAFT
would have taken a fundamentally different approach to exchange them, and the
resulting shapes of the two departments would have been different from those
shown in Figure 6.28. As the above example illustrates, LOGIC can exchange two
unequal-area departments (whether they are adjacent or not). Naturally, other de-
partments have to shift to accommodate such an exchange (We encourage the
reader to compare the layouts shown in Figures 6.26d and 6.28.) Hence, like
MCRAFT, LOGIC can “automatically” shift other departments, when necessary.
However, again like MCRAFT, this may pose a problem if a fixed department is
shifted in the process. One may try excluding all fixed departments from the tree
(to retain their current positions). With such an approach, if a cut goes through
 336
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

60' A
G 75'

40' B
E 50'

75' C
D
F 75'

25' H

80' 120' 160'

Figure 6.28 LOGIC layout obtained after exchanging departments D and E in Figure 6.26d.

one or more fixed departments or obstacles, it will complicate the calculation of

its x - or y - coordinate. With LOGIC, it is generally not straightforward to model
fixed departments or obstacles relative to CRAFT or MIP. We refer the reader to
[62] for details.
LOGIC can be applied in nonrectangular buildings provided that the building
shape is “reasonable.” If a cut intersects a portion of the building where the length
or width changes, then LOGIC uses a simple search strategy to compute the exact
location of the cut. For example, consider the layout shown in Figure 6.29. The lay-
out and the building are identical to the one shown earlier in Figure 6.26d except
for the change we made to the east side of the building. The third cut shown earlier
in Figure 6.27 (i.e., the cut that horizontally divided departments G and {D, F}) has
to be computed differently since the “building” length increases from 130 feet to
200 feet within department G.
In presenting LOGIC, we assumed that the above cuts—and the assignment of
departments to either side of each cut—are made randomly. Actually, in order to

180' 130'

57.14'
66.67' A
G

22.86'
44.44' B

D 60'

66.67' C
E
F 60'
22.22' H

90' 90' 200'

Figure 6.29 LOGIC layout obtained for a nonrectangular building.

 337
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

use LOGIC in a practical application, such decisions are made within the framework
of a pseudorandom search strategy such as “simulated annealing,” which we discuss
in Section 6.6.
We note that layouts obtained by LOGIC are supersets of layouts obtained by
BLOCPLAN and similar algorithms that use “bands” for layout formation. That is,
any layout developed by BLOCPLAN can be expressed as a cut-tree in LOGIC, but
not all cut-trees obtained with LOGIC will yield a layout that is composed of bands.
Therefore, BLOCPLAN’s solution space (and the solution space of other algorithms
that use “bands”) is a subset of LOGIC’s solution space, and consequently one
would expect to obtain lower-cost layouts with LOGIC in general. We also note that
with both LOGIC and BLOCPLAN it is difficult to treat nonfixed departments that
may have prescribed or fixed shapes. Since the final shape of a department depends
on a number of factors that are not known until the cut-tree (or the bands) is con-
structed, there is no easy way to control the final length and width of a nonfixed de-
partment. With few exceptions (such as MIP), many layout algorithms are not very
effective in tackling fixed or prescribed department shapes. We will further discuss
department shapes in Section 6.5.

6.4.8 MULTIPLE
Multifloor Plant Layout Evaluation (MULTIPLE) was developed by Bozer, Meller,
and Erlebacher [7]. As the name suggests, MULTIPLE was originally developed for
multiple-floor facilities, which we address later in this chapter. However, it can also
be used in single-floor facilities simply by setting the number of floors equal to one
and disregarding all the data requirements associated with the lifts.
Except for the exchange procedure and layout formation, MULTIPLE is similar
to CRAFT. It uses a from-to chart as input data for the flow, and the objective func-
tion is identical to that of CRAFT (i.e., a distance-based objective with distances
measured rectilinearly between department centroids). Departments are not re-
stricted to rectangular shapes, and the layout is represented in a discrete fashion.
Also, like CRAFT, MULTIPLE is an improvement-type layout algorithm that starts
with an initial layout specified by the layout planner. Improvements to the layout
are sought through two-way exchanges, and at each iteration, the exchange that
leads to the largest reduction in layout cost is selected; that is, MULTIPLE is a steepest-
descent procedure.
The fundamental difference between CRAFT and MULTIPLE is that MULTIPLE
can exchange any two departments whether they are adjacent or not. Recall that
MCRAFT and BLOCPLAN can also exchange any two departments; however, since
both algorithms are based on “bands,” fixed departments may either shift or change
shape. Also, the band approach imposes certain restrictions on the initial layout as
well as fixed departments and obstacles, as discussed earlier. In essence, MULTIPLE
retains the flexibility of CRAFT while relaxing CRAFT’s constraint imposed on de-
partment exchanges.
MULTIPLE achieves the above task through the use of “spacefilling curves”
(SFCs), which were originally developed by the Italian mathematician Peano. Al-
though SFCs (which were considered “mathematical oddities” at the time of their
introduction) initially had nothing to do with optimization and industrial engineering,
 338
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

2 3

1 4 64

(a) Hilbert curve [26]

3
2 4
4
3 1

1
6 6
5

(b) Layout vector: 1-2-3-4-5-6 (c) Layout vector: 5-2-3-4-1-6

Figure 6.30 MULTIPLE’s use of spacefilling curves in layout formation;

departments 1 and 5 have been exchanged.

they have been used to construct a heuristic procedure for routing and partitioning
problems [5] and for determining efficient locations for items in a storage rack [4]. In
MULTIPLE, SFCs are used to reconstruct a new layout when any two departments
are exchanged.
MULTIPLE’s use of SFCs for the above purpose can perhaps be best described
through an example. Consider the SFC shown in Figure 6.30a, which is known as
the Hilbert curve [26]. (The procedure to generate such a curve is shown in [7]. The
interested reader may also refer to [5].) Note that the curve connects each grid such
that a “dot” traveling along the curve will always visit a grid that is adjacent to its
current grid. Also note that each grid is visited exactly once. Suppose the following
area values (expressed in grids) are given for six departments: A1  16, A2  8, A3 
4, A4  16, A5  8, and A6  12. If the layout vector or the fill sequence is given by
1-2-3-4-5-6, then we obtain the layout shown in Figure 6.30b by starting from grid 1
and assigning the first 16 grids (along the SFC) to department 1, the next eight grids
(along the SFC) to department 2, and so on. In other words, the SFC allows MULTIPLE
to map a unidimensional vector (i.e., the layout vector) into a two-dimensional
 339
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

layout. The above mapping can be performed rapidly since the grids are sorted a
priori according to their sequence on the SFC.
Given such a mapping, it is now straightforward to exchange any two de-
partments by exchanging their locations in the layout vector. For example, to ex-
change departments 1 and 5, which are neither adjacent nor equal in area, we first
switch their positions in the layout vector to obtain 5-2-3-4-1-6 and then simply re-
assign the grids (following the SFC). The resulting layout is shown in Figure 6.30c.
Note that all the departments, except for department 6, have “shifted” to accom-
modate the above exchange. (The shift is fairly significant since department 5 is
only half as large as department 1.) In general, whenever two unequal-area de-
partments are exchanged, all the departments that fall between the two depart-
ments on the layout vector will be shifted; the other departments will remain in
their original locations.
In order to avoid shifting fixed departments or obstacles, the SFC “bypasses”
all the grids assigned to such departments. Also, in those cases where the building
shape is irregular or there are numerous obstacles (including walls), MULTIPLE can
be used with a “hand-generated” curve, which may start at any grid, and end at any
grid, but must visit all the grids exactly once by taking only horizontal or vertical
steps (from one grid to an adjacent grid). Diagonal steps are allowed but generally
not recommended since they may split a department. (Recall that two grids that
“touch” each other only at the corners are not considered adjacent for facility layout
purposes.) Unless a fixed department or wall physically separates the building into
two disjoint segments, it is always possible to construct such a curve by reducing
the grid size. (Why?)
Of course, such hand-generated curves are mathematically no longer SFCs,
but they serve the same function. In [7], the authors report that while using MULTIPLE
in a “large, four-floor production facility,” they opted for hand-generated curves to
capture the “exact building shape, the current layout, and all the obstacles.” In the
following example, we illustrate how a hand-generated curve can be used to capture
the current layout. Such curves are also referred to as “conforming curves” since
they fully conform to the current layout.

Example 6.7
Using MULTIPLE and spacefilling curves to improve a layout
Consider the same data given for Example 6.1, where departments A and G are assumed to
be fixed. The initial layout (which has a cost of 59,480 units) was shown earlier in Figure 6.15.
A hand-generated, conforming curve for the initial layout is shown in Figure 6.31a. Note
that the curve does not visit departments A and G. Also note that the curve visits all the
grids assigned to a particular department before visiting any other department. The initial
layout vector is given by C-B-D-H-F-E.
Given the above initial layout vector and the curve, in the first iteration MULTIPLE ex-
changes departments C and D, which reduces the layout cost to 54,260 units. In the second
and last iteration, MULTIPLE exchanges departments C and H to obtain the layout vector
D-B-H-C-F-E and the final layout shown in Figure 6.31b. The cost of the final layout is equal
to 54,200 units, which is less than the final layout cost obtained by CRAFT (2833.50  20 
56,670 units). In general, MULTIPLE is very likely to obtain lower-cost solutions than CRAFT
because it considers a larger set of possible exchanges at each iteration. However, even if
 340
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
1
2
3
4
5
6
7
8
9
10

(a) Conforming "hand-generated" curve

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
1 A A A A A A A A A A G G G G G G G G
2 A A G G
3 A A A A A A A A A A G G G G
4 D D D D D D D D D D E E G G G G G G
5 D D E E E E E E E E
6 D D D D D D D D D D E E E E E E E E
7 B B B B B B B B B F F F F F F F E E
8 B B B B B B B B B F F F F F F F F F
9 B H H C C C C C C F F F F F F F F F
10 B H H H C C C C C C C C C F F F F F

(b) Final MULTIPLE layout (z = 54,200 units)

A G

D E

F
H C

(c) Final "massaged" layout obtained with MULTIPLE

Figure 6.31 Layouts obtained with MULTIPLE for Example 6.7.

both algorithms are started from the same initial layout, MULTIPLE is not guaranteed to find
a lower-cost layout than CRAFT. (Why?)
As we indicated for CRAFT, the final layout generated by MULTIPLE may also require
massaging to smooth the department borders. Retaining the relative locations of the depart-
ments and slightly adjusting some department areas, we obtain the final layout shown in
Figure 6.31c. Also, note that the particular curve used not only determines the final layout
cost, but also determines the department shapes. Hence, alternative layouts can be gener-
ated by trying different curves.
 341
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

MULTIPLE may also be used as a construction procedure. In such cases, the

layout planner may use any SFC or hand-generated curve that best conforms to the
(vacant) building and possible obstacles; of course, there is no initial layout that
the curve needs to conform to. Generally speaking, curves such as the one shown
in Figure 6.30a seem to generate more reasonable department shapes than those
that “run” in straight lines from one end of the building to the other. Any layout
vector may be used as the “initial” layout. It is generally recommended to try alter-
native layout vectors as the starting point.
It is also instructive to note that, by considering all possible layout vectors for
small problems, one may obtain the “optimal” layout with respect to a given curve.
For example, for the curve we used in Example 6.7, the layout vector D-B-H-C-F-E
is optimal, and the final layout shown in Figure 6.31b (at 54,200 units) is the “opti-
mal” layout. Two alternative curves are shown in Figure 6.32a and b, for which the
optimal layout vector is given by D-E-F-B-C-H (at 54,920 units) and D-E-F-H-B-C (at
54,540 units), respectively. The above results suggest that, while alternative layouts
may be obtained from alternative curves, the cost of these layouts may not be very
sensitive to the curve, provided that the curves are not dramatically different. As an
exercise, the reader may construct the optimal layouts by using the above two lay-
out vectors and the corresponding curves shown in Figure 6.32.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
1
2
3
4
5
6
7
8
9
10

(a)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
1
2
3
4
5
6
7
8
9
10

(b)
Figure 6.32 Alternative curves for MULTIPLE.
 342
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

6.5 DEPARTMENT SHAPES AND MAIN AISLES

In developing alternative block layouts, the main aisles are typically not represented
explicitly until the block layout is finalized. As the layout planner “massages” the
final block layout, he or she attempts to correct irregular department shapes and aims
for smooth department borders primarily for two reasons: first, a poor department
shape may make it virtually impossible to develop an efficient and effective detailed
layout for that department; second, since main aisles connect all the departments,
by definition, irregular department shapes would lead to irregular main aisles. Gen-
erally speaking, for efficient material handling and other reasons (including safety,
unobstructed travel, evacuation in an emergency), main aisles should connect all
the departments in a facility with minimum travel, minimum number of turns, and
minimum “jog overs” (i.e., they should run in straight lines as much as possible).
Hence, attaining “good” department shapes is an important consideration in finaliz-
ing a block layout.
As we showed in Section 6.4 (see Equations 6.38 and 6.39), controlling depart-
ment shapes is relatively straightforward for rectangular departments (such as those
obtained with BLOCPLAN and MIP). This is due to the fact that it is straightforward
to define and measure the shape of a rectangle: it is simply the ratio of its longer
side to its shorter side (or vice versa). However, if the layout planner needs or
wishes to consider nonrectangular department shapes (such as those obtained with
CRAFT and MULTIPLE), then shape measurement and control is not straightforward.
In fact, given two alternative but “similar” shapes for the same department, one of
the alternatives may be regarded as acceptable while the other one is regarded as
poor. Although humans are good at making such (subjective) judgments with
respect to department shapes, computer-based algorithms require formal and objec-
tive measures.
A few alternative measures have been suggested in the literature to control the
shape of nonrectangular departments. As we shall see, such measures are intended
only to detect and avoid irregular department shapes; they cannot be used to guar-
antee or prescribe specific department shapes. Although the shape measures we de-
scribe below are more suitable for the discrete layout representation, they can be
implemented with the continuous representation as well.
Two measures to control department shapes are presented in [38]. Both mea-
sures are based on first identifying the smallest rectangle that fully encloses the
department. The first measure is obtained by dividing the area of the smallest en-
closing rectangle (SER) by the area of the department. Given that the department
area is fixed, one would expect the above ratio to increase as the department shape
becomes more irregular since a larger rectangle would be required to enclose the
department. The second measure is obtained by dividing the longer side of the SER
by its shorter side. As before, as the department shape becomes more irregular, one
would expect the above ratio to increase.
A third measure, presented in [7], originally appeared in papers concerned
with geometric modeling (see [19], among others). It is based on the observation
that, given an object with a fixed area, the perimeter of the object generally in-
creases as its shape becomes more irregular. Hence, one may measure the shape of
a department by dividing its perimeter by its area. However, unless the layout planner
 343
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

generates alternative shapes for each department “by hand” and computes the
above ratio for each shape a priori, it is difficult to predict reasonable values for it.
To address this difficulty, in [7] the above ratio is normalized as follows: if the
“ideal” shape for a department is a square, then the “ideal” shape factor, say, S *, is
equal to (P/A) *  (41A)/A  4/1A, where P denotes the perimeter and A de-
notes the area of the department. The normalized shape factor, say, F, is equal to
S/S *  (P/A)/(4/1A) = P/(4 1A). Hence, if a department is square shaped, we ob-
tain F  1.0; otherwise, we obtain F 1.0. Generally speaking, reasonable shapes
are obtained if 1.0 F 1.4. If a square is not the “ideal” shape for a department,
then the analyst may impose a lower bound greater than 1.0 on F.
In [7], the above three shape factors are compared via an example; we pres-
ent it here with a minor change. Suppose the department area is equal to 16
units. Four alternative shapes, including the SERs, are shown in Figure 6.33. For
Figure 6.33a, the first measure is equal to 1.0, and it increases to 25/16  1.5625
for Figure 6.33b. However, it remains at 1.5625 for Figure 6.33c and d. The sec-
ond measure, on the other hand, is equal to 4.0 for Figure 6.33a, and it decreases
to 1.0 for the remaining three department shapes. The third measure, F, is equal
to 1.25, 1.25, 1.50, and 1.625 for Figure 6.33a through d, respectively. (The
reader may verify that P is equal to 20, 20, 24, and 26 for Figure 6.33a through d,
respectively.)
As the above example illustrates, no shape measure is perfect, although F
seems to generate more accurate results. Furthermore, as we remarked earlier, none
of the shape measures can be used to prescribe a particular shape for a department.
Rather, they can be used to avoid irregular department shapes by checking the
shape factor for each department following an exchange. Since a typical computer

2 5

(a) (b)

5 5

5 5

(c) (d)
Figure 6.33 Alternative department shapes and the smallest enclosing rectangle.
 344
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

run considers many possible department exchanges, any shape measure must be
straightforward to compute. The three shape measures we showed above are
straightforward to implement within a computer-based layout algorithm that uses
the discrete representation; the computational burden they generate is minimal.
For example, consider the third measure. To compute the perimeter of depart-
ment i, the computer first examines each grid assigned to department i, one at a
time, and simply counts the number of adjacent grids which have not been assigned
to department i. (Unless it is located along the building perimeter, each grid has ex-
actly four adjacent grids. Grids that are located along the perimeter of the building
must be treated slightly differently.) Once the perimeter of each department is com-
puted, the computer determines the normalized shape factor for each department.
As a result of exchanging two departments, if any department violates a shape con-
straint (i.e., if the normalized shape factor for the department falls outside a user-
defined range, which may be department-specific), then the computer simply “rejects”
the exchange.
Note that, with the above scheme to compute department perimeters, those with
“enclosed voids” will be correctly identified as possibly irregular in shape. For example,
for the departments shown in Figure 6.9c, d and e, the normalized shape factor is given
by 1.061, 1.591, and 1.414, respectively. Had we not included the “inside perimeter” in
Figure 6.9e, the shape factor would have been 1.061, which is a misleading figure.
Although shape measures are useful in avoiding irregular department shapes,
they should be used with care for three reasons: First, if strict shape constraints are im-
posed, many (if not all) possible department exchanges may be rejected, regardless of
how much they reduce the layout cost. Second, some departments may take irregular
shapes only temporarily; that is, a department which is irregularly shaped in the cur-
rent iteration may assume a reasonable shape in the next iteration—note that this may
occur with both CRAFT and MULTIPLE. Third, the analyst has the option to correct the
department shapes by massaging the final layout generated by the computer. If strict
shape constraints are imposed, the computer will “automatically” discard lower-cost
solutions with irregular department shapes even if the analyst could have corrected the
problem through massaging. Hence, we generally recommend imposing no shape
constraints until the layout planner makes a few runs and obtains preliminary results.
In subsequent runs, a shape constraint may be imposed only on those departments
that have unreasonable shapes that cannot be corrected through massaging.

6.6 SIMULATED ANNEALING AND GENETIC

ALGORITHMS3
Simulated annealing (SA) and genetic algorithms (GAs) represent relatively new
concepts in optimization. Although both SA and GAs can be used for layout con-
struction, we will limit their presentation to layout improvement. Since a formal and

3
This section presents research results without examples. Intended for first-year graduate students and ad-
vanced undergraduate students, it provides some insight into the use of “metaheuristics” in layout design.
 345
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

thorough treatment of either subject is beyond the scope of this book, we will present
only some of the basic concepts of SA and GAs, and show only one SA-based layout
algorithm.
The most significant drawback of the steepest-descent approach (which we
described in Section 6.4) is that it forces the algorithm to terminate the search at the
first two-opt or three-opt solution it encounters. As we remarked earlier, such a solu-
tion is very likely to be only locally optimal. (Note that, when they consider two-
way or three-way department exchanges, CRAFT, BLOCPLAN, and MULTIPLE are
examining the local “neighborhood” of the current solution.) As a result, any algo-
rithm that uses the steepest-descent technique becomes highly “path dependent” in
that the initial solution and the specific department exchanges made by the algo-
rithm play a significant role in determining the cost of the final solution. (The im-
pact of the initial solution is also known as the initial layout bias.) Ideally, a
heuristic procedure should consistently identify a low-cost, that is, an optimal or
near optimal, solution regardless of the starting point.
One of the primary strengths of SA is that, while trying to improve a layout, it
may “occasionally” accept nonimproving solutions to allow, in effect, the algorithm
to explore other regions of the solution space (instead of stopping at the first “seem-
ingly good” solution it encounters). In fact, an SA-based procedure may accept non-
improving solutions several times during the search in order to “push” the algorithm
out of a solution that may be only locally optimal. As a result, the objective function
value may actually increase more than once. However, the amount of increase in
the objective function that the algorithm will “tolerate” is carefully controlled
throughout the search. Also, the algorithm always “remembers” the best solution;
that is, the best solution identified since the start of the search is never discarded
even as new regions of the solution space are explored.
The fundamental concepts behind SA are based on an interesting analogy be-
tween statistical mechanics and combinatorial optimization problems [34]. Statistical
mechanics is the “central discipline of condensed matter physics, a body of methods
for analyzing aggregate properties of the larger numbers of atoms to be found in
samples of liquid or solid matter” [34]. One of the key issues in statistical mechanics
is the state of the matter (or the arrangement of its atoms) as its temperature is grad-
ually reduced until it reaches the “ground state” (which is also referred to as the
“lowest energy state” or the “freezing point”).
According to [73], “In practice, experiments designed to find the (lowest en-
ergy) states are performed by careful annealing, that is, by first melting the (ma-
terial) at a high temperature, then lowering the temperature slowly (according to
an annealing schedule), finally spending a long time at temperatures in the vicin-
ity of freezing, or solidification, point. The amount of time spent at each temper-
ature during the annealing process must be sufficiently long to allow the system
to reach thermal equilibrium (steady state). If care is not taken in adhering to the
annealing schedule (the combination of a set of temperatures and length of time
to maintain the system at each temperature), undesirable random fluctuations
may be frozen into the material thereby making the attainment of the ground
state impossible.”
Hence, the analogy between combinatorial optimization problems and statis-
tical mechanics is that each solution in the former corresponds to a particular
arrangement of the atoms in the latter. The objective function value is viewed as the
 346
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

energy of the material, which implies that finding a low-cost solution is analogous
to achieving the lowest energy state through an effective annealing schedule. Also,
since we are working with mathematical optimization models rather than “con-
densed matters,” the annealing schedule applied to a combinatorial optimization
problem is not “real” annealing, but “simulated” annealing. As we shall see shortly,
the annealing schedule, that is, the set of temperatures used (including the initial
temperature) and the time spent at each temperature, plays an important role in al-
gorithm development and the cost of the final solution.
The concept of “occasionally” accepting nonimproving solutions goes back to
a simple Monte Carlo experiment developed in [44]: Given a current arrangement of
the atoms (or “elements”), we randomly make incremental changes to the current
arrangement to obtain a new arrangement, and we measure the decrease in energy,
say, E. If E 0 (i.e., if the energy decreases), then we accept the new arrange-
ment as the current one and use it to make subsequent changes. However, if E
0, then the new arrangement is accepted with probability P( E )  exp( E/kbT ),
where T is the temperature and kb is Boltzmann’s constant. (If we do not accept the
new arrangement, then we generate another one.)
To apply the above procedure to optimization problems, we simply treat
the “current arrangement” as the current solution, the “new arrangement” as the
candidate solution, and the “energy” as the objective function value. A random
incremental change is made, for example, by exchanging two randomly picked
departments in the current layout. Last, we set kb  1 since it has no known sig-
nificance in optimization problems. Note that, as the increase in the objective
function gets larger, the probability of accepting the candidate solution gets
smaller. For example, if T  150 and the increase in the objective function is
equal to 250 units (i.e., if E  250), then the probability of accepting the can-
didate solution is equal to 0.1889. At T  150, if the increase in the objective
function is equal to 400 units, however, then the above probability decreases to
0.0695. Also note that the probability of accepting a nonimproving solution de-
creases as the temperature decreases. For example, for E  250, if we decreased
the temperature from 150 to 100 units, then the above probability would decrease
from 0.1889 to 0.0821. In other words, we are more likely to accept nonimproving
solutions early in the annealing process (due to the relatively high temperatures).
It is instructive to further note that the probability in question ultimately de-
pends on the change in the objective function relative to the temperature. There-
fore, in an SA-based algorithm, we are not only concerned with how fast we “cool”
the system but also with the initial temperature we select to start the annealing
process. One possible approach is to set the initial temperature according to the
objective function value of the starting solution (see, for example, [31]). In the algo-
rithm that follows, the initial temperature is set equal to z0/40, where z0 is the cost
of the initial layout. Of course, the layout planner should experiment with different
initial temperature settings.
The above process of generating candidate solutions and appropriately updat-
ing the current solution continues until the system reaches steady state at the cur-
rent temperature. As shown in the following algorithm, the percent change in the
mean objective function value is used to determine whether steady state has been
reached. According to [73], “at each temperature, the annealing schedule must allow
 347
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

the simulation to proceed long enough for the system to reach steady state.” Once
steady state is reached, the temperature is reduced according to a predetermined
temperature reduction factor, and we continue to generate and evaluate candidate
solutions with the new temperature setting.
Typically, the search is terminated either when a user-specified final temper-
ature is reached (which can also be expressed as the maximum number of temper-
ature reductions to be considered) or a user-specified number of successive
temperature reductions does not produce an improvement in the best solution iden-
tified since the start of the annealing process (i.e, the “current best” solution). The
algorithm we present below uses both the latter stopping criterion and one that is
based on the maximum number of “epochs.” An epoch corresponds to a particular
set of candidate solutions accepted by the algorithm. The epoch length is expressed
as the number of candidate solutions in the set.
To develop a SA-based layout algorithm, we will apply an annealing schedule
to MULTIPLE, which was described in Section 6.4. Recall that, in MULTIPLE, given a
particular spacefilling curve and the departmental area requirements, the layout
vector (i.e., a sequence of department numbers) fully defines a layout. Therefore,
any solution is represented only as a layout vector. Except for the technique used in
generating the candidate layout vectors, the algorithm we show here is identical to
Simulated Annealing-Based Layout Evaluation (SABLE), which was developed by
Meller and Bozer [41]. (SABLE uses a more general technique for generating candi-
date layout vectors.) When appropriate, variables were used in a manner similar to
their use in [73]. Let

s0 be the initial layout vector,

s* be the “current best” layout vector, which corresponds to the lowest cost
layout identified by the algorithm,
s be the current layout vector,
s be the candidate layout vector,
 be the temperature reduction factor (which controls how fast the system
is “cooled down”),
T be a set of annealing schedule temperatures {t1, t2, t3, …}, where ti 
t0()i for all i 1,
t0 be the initial temperature,
e be the (fixed) epoch length,
fj (s) be the objective value of the j th accepted candidate layout vector, s, in
an epoch,
e
fe be the mean objective function value of an epoch 3i.e., fe = 3g j = 1 fj (s)4>e4
f e be the overall mean objective function value of all the layout vectors
accepted during the epochs previous to the current one (for a given
temperature),
Hi be a threshold value used to determine whether the system is in equilib-
rium at temperature i,
 348
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

M be the maximum total number of epochs to be considered (across all

temperatures),
I be a counter to record the last temperature setting that produced the
“current best” layout vector, s *,
N be the maximum number of successive temperature reductions that will
be performed with no improvement in s*.
The initial layout vector s0 as well as the values of , t0, e, H, M, and N are spec-
ified by the user a priori. Using the above notation (except for the subscript j, which
we will omit for brevity), a simple version of SABLE [41] is presented as follows:

Step 1. Set s  s0, I  1, and i  1. Compute the initial layout cost, z0; set t0 
(z0)/40 and t1  t0.
Step 2. For the current layout vector, compute the layout cost f (s).
Step 3a. Randomly pick two departments in s, exchange their locations, and store
the resulting layout vector (i.e., the candidate layout vector) in s.
Step 3b. Compute the decrease in the layout cost (i.e., set f  f (s)  f (s)). If f
0, go to step 3d; otherwise, go to step 3c.
Step 3c. Sample a random variable x ' U (0, 1). If x  exp( f/ti), go to step 3d;
otherwise, go to step 3a.
Step 3d. Accept the candidate sequence; that is, set s  s and f (s)  f (s). If
f (s)  f (s *), then update the “current best” solution; that is, set s *  s,
f (s *), f (s), and I  i. If e candidate sequences have been accepted, go
to step 4; otherwise, go to step 3a.
Step 4. If equilibrium has not been reached at temperature ti —that is, if
|fe - fe ¿ |/ fe ¿ Ú Hi—reset the counter for accepted candidate solutions
and go to step 3a; otherwise, set i  i  1 and ti  t0()i. If (i  I )  N,
go to step 5; otherwise, STOP—the maximum number of successive
nonimproving temperatures has been reached.
Step 5. If the total number of epochs is less than M, go to step 3a; otherwise,
STOP.
The parameter values selected by the user are very likely to have a more-than-
minor impact on the cost of the final solution obtained by the above algorithm as
well as its execution time. Setting the initial temperature with respect to the initial
layout cost and ensuring that the system is not “cooled down” too rapidly generally
improves the performance of the algorithm. In [41], for example, the following
parameter values were observed empirically to substantially reduce the initial layout
bias and yield generally “good” solutions for all the test problems evaluated:  
0.80, t0  (z0)/40, e  30, H1  0.25, Hi  0.05 for all i 2, M  37, and N  5. We
caution the reader that the above parameter settings are shown only as an example.
(Also, recall that in [41] a more general candidate solution generation procedure is
used.) Once an SA-based algorithm is developed, it is generally good practice to ex-
periment with a variety of parameter settings.
The above algorithm is certainly not the only possible application of SA to
facility layout problems. In fact, as we remarked in Section 6.4, LOGIC [62] is actually
 349
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

an SA-based layout algorithm. Recall that LOGIC develops a layout by executing a

series of horizontal and vertical cuts, and by assigning (an appropriate subset of)
the departments to one or the other side of each cut. Given a current solution,
which is represented by a cut-tree (see Section 6.4), a candidate solution may be
obtained, for example, by randomly changing the orientation of one (or more) of
the cuts in the tree and/or by randomly changing the partition of a set of depart-
ments over one (or more) of the cuts. (We refer the reader to [62] for further details.)
Candidate solutions obtained in the above manner can be evaluated within the
framework of an annealing schedule as we showed in the above algorithm. Another
application of SA in facility layout is presented in [30]. The algorithm in [30] is in-
tended primarily for the special case where all the departments have equal area re-
quirements; using it with general departmental area requirements is possible but it
may lead to numerical problems or split departments.
The “biased sampling technique” (BST) [50], which was applied to CRAFT,
has certain similarities to SA. With the BST, at each iteration, given a set of ex-
changes (that are estimated to reduce the layout cost), instead of always selecting
the exchange that is estimated to reduce the layout cost the most, the algorithm as-
signs a nonzero probability to each exchange in the set and randomly selects one
of them. By solving the same problem several times with different random number
streams and different starting points, the BST was shown to reduce the initial lay-
out bias [50]. However, the BST is not as “formal” a concept as SA, and, more im-
portantly, the BST will still terminate at the first locally optimal solution it
encounters (i.e., the BST does not accept an exchange that may increase the objec-
tive function value).
It is instructive to note that developing effective and efficient SA-based lay-
out algorithms is not only a matter of finding a “good” annealing schedule but
also a matter of finding a “good” representation that makes it possible to rapidly
generate and evaluate a variety of candidate solutions. Note that we are referring
to “solution” representation and not “layout” representation (as in discrete versus
continuous layout representation). Naturally, the two are not independent. For
example, in MULTIPLE [7] and SABLE [41], the solution is represented as a layout
vector, and the layout is constructed through the use of spacefilling curves
(which work well with the discrete layout representation). In LOGIC [62] the so-
lution is represented as a cut-tree, and the layout is constructed by applying ver-
tical and horizontal cuts (which work well with the continuous layout
representation).
The application of GAs, on the other hand, to facility layout problems (and
other optimization problems) is relatively recent, but it has been gaining momen-
tum. The basic concept behind GAs was developed by Holland [27] who observed
that the “survival of the fittest” (SOF) principle in nature may be used in solving
decision-making, optimization, and machine-learning problems. Although the SOF
principle may first appear to have no relation to optimization problems, the “rela-
tionship” between the two is as fascinating as the “relationship” we described be-
tween statistical mechanics and optimization problems.
Algorithms based on SA “occasionally” accept nonimproving solutions; how-
ever, they still work with only one solution at a time. That is, there is only one
current solution, from which we generate only one candidate solution. In contrast,
 350
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

GAs work with a family of solutions (known as the “current population”) from
which we obtain the “next generation” of solutions. When the algorithm is used
properly, we obtain progressively better solutions from one generation to the
next. That is, good solutions (or actually “parts of good solutions”) propagate
from one generation to the next and lead to better solutions as we produce more
generations.
The basic GA, at least on the surface, is admirably simple. Given a current
population of, say, N solutions, we generate the next population as follows: We
randomly pick two “parents” (i.e., two solutions) from the current population
and “randomly cross over” the two parents to obtain two “offspring” (i.e., two
“new” solutions) for the next population. We repeat the above process until we
have a new generation with N solutions. (Note that, unlike most natural systems,
the population size is fixed at N.) Each time we generate two offspring, their two
parents are picked randomly; however, the probability of picking a particular so-
lution as a parent is proportional to the “fitness” of the parent. For minimization
problems, the above implies that the probability of picking a solution as a parent
is inversely proportional to the objective function value of that parent (i.e., a “fit”
parent with a small objective function value is more likely to be selected). The
first generation (i.e., the starting generation) is usually created randomly. Typi-
cally, the process of creating new generations continues until a prespecified
number of generations have been produced or no noticeable reduction in the av-
erage population objective function value is detected for a number of successive
generations.
In most GAs, the above “basic” algorithm is modified in a number of ways.
One common addition is “mutation,” which essentially means altering (in some ar-
bitrary fashion) one or more solutions picked at random within a given popula-
tion. Another addition is the concept of “elitist reproduction,” which basically
implies that when a new population is created, the best 10 or 20% of the solutions
in the current population are “automatically” copied over to the next population.
The remaining 90 or 80% of the solutions in the next population are generated by
the “two parents–two offspring” method we described above. Solutions that were
copied over are still eligible to act as parents for creating the remainder of the
next population.
The population size, the mutation rate, the rate of elitist reproduction, and the
number of generations to create are all user-specified parameters that affect the per-
formance and execution time of a GA. Also, there are alternative cross-over opera-
tors that specify how two offspring are “randomly” generated from the two parents.
In fact, the type of solution representation used for the problem is critical in terms
of defining appropriate cross-over operators (so that we do not create “infeasible off-
spring” from “feasible parents”). That is, even though the two parents represent fea-
sible solutions, if we do not use or develop an appropriate cross-over operator, their
offspring may not. A discussion of alternative cross-over operators and appropriate
values to assign to the above parameters are beyond the scope of this book. The
reader may refer to [21] for an excellent introduction to basic and some advanced
concepts in GAs and their use in optimization problems and machine learning. Also,
[10] is a good reference book for GAs; it includes comprehensive examples for GA
applications. Furthermore, the reader will find GAs applied to the quadratic assign-
ment problem and the facility layout problem in [63], [65], and [66], among others.
 351
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

In concluding this section, we note that, generally speaking, if the computa-

tional effort required to evaluate a single solution is high, then SA-based algorithms
are likely to outperform GAs. This stems primarily from the fact that a GA is more
likely to evaluate a larger number of solutions than is an SA-based algorithm. It may
also be the case that, in early generations, GAs spend computer time evaluating
many poor solutions before they are removed from the “gene pool.” On the other
hand, there is a “natural fit” between GAs and parallel computers, where each
processor in the computer can independently and concurrently generate and evalu-
ate a pair of offspring for the new population. This way, a large number of new pop-
ulation members can be rapidly generated and evaluated in parallel. Nevertheless,
with or without parallel computers, it is too early to make conclusive statements
about SA versus GAs in facility layout; further research is needed to fully develop
both techniques and compare their performance on various facility layout problems.

6.7 MULTIFLOOR FACILITY LAYOUT

With some exceptions such as offices, it is perhaps safe to say that most industrial
facilities in the United States are single-floor facilities. This is especially true for
manufacturing, warehousing, and distribution facilities. However, many such facil-
ities also use mezzanines, which can be modeled as a “partial second floor” for
layout purposes. (More specifically, a mezzanine is modeled as a full second floor
with an obstacle. Recall that an obstacle is essentially a fixed department with zero
flow. The obstacle is placed on the portion of the second floor that falls outside the
mezzanine.)
Furthermore, in addition to some older, multifloor industrial facilities still in
use in the United States, a large number of existing and new industrial facilities in
countries such as Japan, Korea, and China are multifloor facilities since usable
land is either very limited and/or very expensive, especially as one gets closer to
industrialized zones. Although a multifloor facility presents certain challenges
(see, for example, Schonberger [55], pp. 120–121), many manufacturers in such coun-
tries prefer a multifloor facility because it saves them considerable capital, and
they depend on the infrastructure (energy, highways, railways, etc.) that is often
lacking in rural areas where land may be readily available.
Generally speaking, a multifloor facility layout problem is more challenging
than a comparable single-floor layout problem. The incremental complexity is
mostly due to vertical travel between floors. While advances in material handling
technology (to handle material between floors) provide the layout analyst with a
number of options (see, for example, [69] and [70]), the number and location of ver-
tical handling devices to use, the congestion and delays they may create, and the
possible lack of coordination between departments on separate floors are important
factors that will impact the overall quality and effectiveness of any layout developed
for a multifloor facility. Moreover, space constraints in multifloor facilities may be
more limiting. For example, due to floor-loading capacities, floor-to-ceiling clear
height, heat generation, chemical processes involved, and so forth, certain depart-
ments may be restricted to certain floors.
 352
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

Also, even if the total usable floor space available in the building exceeds the
sum of the floor-space requirements of the departments, not all layouts will be fea-
sible. Consider, for example, a three-floor facility with 9 unit squares of available
floor space per floor. Suppose seven departments (numbered 1 through 7) have
the following area requirements: 4 unit squares each for departments 1 through
5 and 3 unit squares each for departments 6 and 7, for a total of 26 ( 27) unit
squares. Although there is excess space in the building, there is no feasible assign-
ment of departments to floors. Therefore, unless the departments are redefined,
there is no feasible layout. Recall that splitting a department, within or across
floors, is not allowed.
Of course, the above small example is a special case, but the main point is that
department-to-floor assignments play an important role in defining the layout alter-
natives and eventually the layout cost. Although a number of layout algorithms have
been developed for the multifloor facility layout problem (see, for example, BLOC-
PLAN [12], MSLP [33], SPACECRAFT [32], and SPS [38]), they have certain shortcom-
ings such as splitting departments across two or more floors, not considering any
lifts or allowing only a single lift (or single bank of lifts) for vertical handling, and/or
assuming equal-area departments. Two algorithms that work with multiple floors,
multiple lift locations, and unequal-area departments (without splitting them across
floors) are MULTIPLE [7] and SABLE [41], which we described earlier. Since SABLE
was shown to be more effective than MULTIPLE, here we will briefly describe
SABLE’s implementation in multifloor facilities.
Recall that SABLE is a simulated annealing-based layout-improvement algo-
rithm. We presented a simplified, single-floor version of SABLE in Section 6.6, where
a candidate layout vector is generated by exchanging the locations of two depart-
ments in the current layout vector. Given a candidate layout vector, SABLE starts with
the first department in the vector and places it in the first floor. The second depart-
ment in the vector is also placed in the first floor and so on, until the placement of
the next department in the vector would create a space overflow in the first floor. At
that point, the next department is placed in the second floor, and the procedure con-
tinues until either all departments in the vector are successfully placed or the candi-
date layout vector is declared infeasible (in which case a new candidate layout vector
is generated from the current vector again by exchanging the locations of two de-
partments in the current layout vector). Each department is placed according to a
spacefilling curve defined for that floor; that is, each floor has its own spacefilling
curve. The number and location of the lifts are specified by the analyst.
Once a feasible candidate layout vector is generated and all the departments
are placed, the resulting multifloor layout is evaluated by the following expression:

min z  a a (cijHdijH + cVij dVij )fij , (6.42)

i j

where fij is the flow from department i to department j (expressed in number of unit
loads moved per unit time), cijH (cijV ) is the cost of, or the relative “weight” associated
with, moving a unit load one horizontal (one vertical) distance unit from depart-
ment i to department j, and d ijH (dijV ) is the horizontal (vertical) distance from depart-
ment i to department j. Note that Equation 6.42 is very similar to the single-floor,
distance-based layout objective. The horizontal distance from department i to j, dijH,
is assumed to go through the lift that minimizes the total horizontal rectilinear
 353
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

distance between the two department centroids. Of course, the vertical distance, dijV,
is measured between the floors. Lift congestion and the throughput capacity of the
lift(s) are not taken into account.
Based on the value of the objective function as determined by Equation 6.42,
the annealing procedure proceeds as described earlier (see step 3b and subsequent
steps in Section 6.6), and the algorithm stops when one of the stopping criteria is
encountered.
A somewhat different approach to multifloor layout is to solve the problem in
two stages. In the first stage, each department is permanently assigned to one of the
floors. In the second stage, the layout of each floor is improved using simulated an-
nealing. Of course, in the first stage, the objective is to minimize the vertical compo-
nent of the objective function given by Equation 6.42, and in the second stage, the
objective is to minimize the horizontal component. Generally speaking, solving a
problem in two stages is not as effective as solving it as a single-stage problem even
if a good or optimum solution is found for each of the two stages. However, the
above two-stage approach is appealing because minimizing vertical travel (by plac-
ing two departments with high interdepartmental flow within the same floor as
much as possible) has managerial appeal, and it also reduces the solution space we
need to explore when we tackle the second stage.
A two-stage multifloor layout algorithm, STAGES, is presented in [42], where
the first stage is solved via a linear mixed integer programming (MIP) model assum-
ing that the inter-floor distances are equal (i.e., the distance between floors 1 and 2
is equal to the distance between floors 2 and 3, and so on). Otherwise, the first
stage results in a quadratic assignment problem (QAP), which has a nonlinear ob-
jective. (The QAP is described in Chapter 10.) Although there are many heuristic
procedures developed for the QAP, it is generally a difficult problem to solve.
Therefore, even if the inter-floor distances may not be exactly equal, using the MIP
model presented below would probably be the preferable approach in most cases
to obtain a solution to stage 1.
Let d gh
ij denote the vertical distance from department i to department j if de-
partment i is assigned to floor g and department j is assigned to floor h. Assuming
that the inter-floor distance between any two adjacent floors is equal to d, we have
ij  d|g  h|, where i, j  1, …, N and g, h  1, …, G. Suppose our (binary) de-
d gh
cision variable is denoted by xig such that xig is equal to 1 if department i is assigned
to floor g and 0 otherwise. If we let yi denote the floor number of department i, that
is, yi = a Gg= 1 gxig , then we obtain d ijgh  d|yi  yj|. Note that the variable yi is an in-
teger variable, but we do not declare it as an integer variable, because it “automati-
cally” assumes integer values as long as the xig values are restricted to 0 or 1.
Suppose the set of positive flows is denoted by F  {fij }. That is, fij 0 for
all i, j H F. Let|F|  M; that is, let M denote the number of positive flows in the
ij denote the mth positive flow (m  1, …, M). The linear
flow matrix. Also, let f m
MIP model to assign departments to floors with the objective of minimizing total
vertical travel (that is, the vertical component in Equation 6.42) is given as
follows:

M
Minimize z = a Vm (6.43)
m=1
 354
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

G
Subject to yi = a gxig for i  1, …, N (6.44)
g=1

Vm Ú (cVij f ijm )(yi - yj ) for m  1, …, M (6.45)

Vm Ú (cVij f ijm )(yj - yi ) for m  1, …, M (6.46)

G

a xig = 1 for i  1, …, N (6.47)

g=1

a aixig … Ag for g  1, …, G (6.48)

i=1

where ai is the floor space required by department i and Ag is the available floor
space on floor g. Constraint set 6.44 determines the floor number of department i
based on the value of the binary variables. Constraint sets 6.45 and 6.46 together de-
termine the flow times vertical distance for the mth flow. Constraint set 6.47 ensures
that each department is assigned to exactly one floor. Constraint set 6.48 ensures
that the total available floor space on each floor is not exceeded. Recall that xig is a
binary variable.
Once the department-to-floor assignments are determined by solving the
above linear MIP model, the floor assignment for each department is fixed and the
second stage of the algorithm is executed. In STAGES, a modified version of SABLE
is used to tackle the second stage. That is, the current layout vector is perturbed to
generate a candidate layout vector without changing the floor assignments of the
departments, and a spacefilling curve (entered for each floor) is used to re–lay out
the affected floors. For the simplified version of SABLE we presented in Section 6.6,
the second stage of STAGES can be implemented simply by randomly picking a
floor and exchanging the location of two randomly picked departments in the lay-
out vector of that floor. (A slightly different and more general approach is used in
[42] to perturb the current layout vector.)
According to the numerical results presented in [42], the layout costs obtained
from STAGES are, in general, better than those obtained from SABLE and better than
another annealing-based algorithm where the department-to-floor assignments
obtained in the first stage are allowed to change as the second stage is solved. We
refer the interested reader to [42] for further details.

6.8 COMMERCIAL FACILITY LAYOUT PACKAGES

The layout algorithms in this chapter are shown with the intent to familiarize the
reader with how different models generate, evaluate, and/or optimize the layout of
a facility. As such, the algorithms we covered are primarily research-based algo-
rithms, and most of them are unfortunately not available as commercial layout de-
sign packages. They also do not perform some of the common analyses performed in
industry, such as spaghetti charting, which can be a helpful visual aid in evaluating
 355
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

alternative layouts. However, there are a number of commercially available layout

design packages that will evaluate/optimize a layout as well as prepare spaghetti
charts and compute metrics such as parts travel (which is the term used for the dis-
tance-based objective function we presented earlier as Equation 6.1).
In past editions of this book, we provided brief descriptions of some of the
commercially available layout design packages only to find that the market
shifted often, with some packages changing ownership and others being, modi-
fied, expanded, or pulled off the market. With that in mind, in this edition, we
only provide an alphabetical list of some of the commercially available layout de-
sign packages as follows:
Flow Planner by Proplanner
www.proplanner.com/product/details/flowpath.aspx
Layout-iQ™ by Rapid Modeling Corporation
http://www.layout-iq.com/
Plant Design & Optimization by Siemens PLM Software
Suite of software: FactoryCAD, FactoryFlow, FactoryMockup, and Plant
Simulation
http://www.plm.automation.siemens.com/en_us/products/tecnomatix/
plant_design/index.shtml
VIP-PLANOPT by Engineering Optimization Software
http://www.planopt.com/VIP-PLANOPT/
Although we provided the URL for each package, such information is very
likely to change. We therefore encourage the reader to use the Web to keep abreast
of new developments. Trade/professional societies such as IIE (Institute of Indus-
trial Engineers) periodically publish lists of software packages for facilities plan-
ning/design or simulation, which are also an excellent source of information for the
layout analyst/engineer.

6.9 THE IMPACT OF CHANGE

The need for a facility layout study can arise under a variety of circumstances. For
example, some of the more common situations that arise in the context of plant lay-
out include the following:
1. Changes in the design of an existing product, the elimination of products from
the product line/family, and the introduction of new products
2. Changes in the processing sequences for existing products, replacements of
existing processing equipment, and changes in the use of general-purpose
and special-purpose equipment
3. Changes in production quantities and associated production schedules, result-
ing in the need for capacity changes
4. Changes in the organizational structure as well as changes in management
philosophies concerning production strategies such as the adoption of lean
concepts, total quality management, etc.
 356
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

5. Incremental changes needed in the layout due to continuous improvement

and kaizen events
If requirements change frequently, then it is desirable to plan for change and to de-
velop a flexible layout—one that can be easily modified, expanded, or contracted.
In addressing the impact of change on facilities planning for electronics manufactur-
ing, Propst [52, p. 16] noted:
Our ancestors could deal with changes as an evolutionary factor. Change could be
digested in small steps or ignored for a lifetime. Today, it is the dominating reality;
our new natural state of affairs.
Curiously, it is the lack of flexibility in our physical facilities that is proving to
be the bottleneck in electronics production. A great many irritations stem from ser-
vices and facilities that respond too slowly, or not at all, to our buildings, furnishings,
and service that have to be revitalized and revisualized.

Flexibility can be achieved by utilizing modular office equipment, workstations,

and material handling equipment; installing general-purpose production equipment;
utilizing a grid-based utilities and services system; and using modular construction.
Additionally, the design of the facility can have a significant impact on the ease and
cost of expansion. We will discuss some of these issues in more detail below.

6.9.1 Adapting to Change and Planning

for Facility Reorganization
Before getting too specific about how to plan for change, it would be worthwhile to
step back and observe that in many manufacturing organizations there are cycles of
expansion and decline due to the very nature of the business environment. In other
words, the manufacturing environment is very dynamic. The facility layout should
also be treated as dynamic. In as much as businesses should have long-term busi-
ness strategies, we must also have a multiyear master plan for facility layout. This
master plan should be consistent with the company’s business plan, and it should
attempt to anticipate future requirements and make provisions for adapting to
changes in facility requirements. Often, prior decisions impose severe constraints
that make it very difficult to institute innovative changes in the facility layout. A few
examples include placements of shipping and receiving docks, locations of heavy
machineries, clear building heights, floors with low load-bearing capacities, load-
bearing walls, location of utilities, and so on.
The master layout plan must also provide the means for a facility to react
quickly to change, adding capacity in a short period of time, or to be able to operate
efficiently at scaled-down operating levels. The facility design must be flexible in
order to provide this high level of responsiveness. We note, however, that varia-
tions in production requirements should not be interpreted as signaling a need to
change the facility layout. Often, opportunities for adjustments can be made by
seeking more efficient machine schedules, better maintenance of equipment,
smoother material flows, and closer coordination with customers and suppliers in
identifying the criticality of delivery dates. When all else fails, it may be time to do
a new layout of the production facility. Or in some cases, it may be the easiest al-
ternative to implement, particularly if the cost of change is marginal. And when
 357
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

these occasions arise, the layout must be flexible enough to quickly accommodate
these changes.
How do we develop such a flexible layout? Harmon and Peterson [22] suggest
the use of the following objectives:
1. Reorganize factory subplants to achieve superior manufacturing status.
2. Provide maximum perimeter access for receiving and shipping materials, com-
ponents, and products as close to each subplant as practical.
3. Cluster all subplants dedicated to a product or product family around the
final process subplant to minimize inventories and shortages, and improve
communication.
4. Locate supplier subplants of common component subplants in a central loca-
tion to minimize component travel distances.
5. Minimize the factory size to avoid wasted time and motion of workers.
6. Eliminate centralized storage of purchased materials, components, and assem-
blies and move storage to focused subplants.
7. Minimize the amount of factory reorganization that will be made necessary by
future growth and change.
8. Avoid locating offices and support services on factory perimeters.
9. Minimize the ratio of aisle space to production process space.
The idea of breaking up a factory into smaller entrepreneurial units is not a
new concept (see Skinner [61]) Each subunit, or a subplant, can be organized much
more efficiently. Alternative layout configurations can be designed for each sub-
plant to take advantage of their specific product and process requirements. An il-
lustration of a factory organized along the subplant concept is given in Figure 6.34.
The illustration shows a major flow structure revolving around a spine flow and
material flows within each subplant organized along U-shaped, I-shaped, and
S-shaped flow structures4 (see Figure 6.34a). You will note that facility expansions
and contractions are easily achieved with this modular configuration by simply ex-
tending outward or contracting inward within each subunit, as shown in Figure 6.34b.
A critical consideration, however, is the central flow structure, which could end
up as the bottleneck when the total material handling requirements between
subplants exceeds the capacity limits of the spine flow server. Sufficient capacity
must be incorporated in the initial design in anticipation of future material flow
requirements.

6.9.2 Volvo Facilities

Volvo Skovdeverken’s factory for the manufacture of four-cylinder engines is de-
signed on the basis of the modular concept. According to Volvo, the factory
is characterized by a new concept of layout, environment, technology and work or-
ganization. The company’s own technicians have worked together with representatives

4
For a comprehensive discussion on alternative material flow structures, see Tanchoco [64].
 358
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

A B A B
Shipping/ Shipping/
receiving receiving

C D E C D E

(a) Spine layout with U, I, and S flows (b) Layout after expanding and contracting
within departments some departments

Figure 6.34 Illustration of a flexible layout.

of the employees and with outside experts during the planning and execution. This
has resulted in a further development of the striving for a better working environment
worker satisfaction and well-being which has been typical for Volvo’s activities in var-
ious places for many years.
Since practical considerations require that the machining, assembly and test de-
partments shall adjoin each other the factory layout is of basic importance. The objec-
tive was to create the atmosphere of a small workshop while retaining the advantages
of rational production and a flexible flow of materials that a large factory makes pos-
sible. The result is a layout quite different from the traditional pattern and giving
many practical and environmental advantages.
The new factory consists of a body containing the assembly and test depart-
ments with four arms at right angles which contain the machining departments. These
latter are separated from one another by generous garden areas. The method of as-
sembly of the engine is one of the most interesting features, both from the technical
and the work organizations points of view. The conventional assembly line principle
has been replaced by an extremely flexible system of group assemblies. Electrically
driven assembly trucks (AGVS) which are controlled by the assembly personnel, to-
gether with other technically advanced solutions, have given the work of building the
engine a new interest. The different teams, in the machining departments as well as
the assembly, have cooperated in the design of the workplaces and are now taking
part in the development of new forms of work organization [71].

As shown in Figure 6.35, the four manufacturing modules are connected

by a material flow/personnel flow spine at each end of the module. The plant is
approximately 400,000 ft2 in size. Windows in each manufacturing module and
break area overlook landscaped gardens.
Volvo Kalmarverken’s automobile assembly plant provides another example
of a modular design. As shown in Figure 6.36, the assembly plant consists of four
equal-sized hexagonal modules, with three two-story assembly modules and one
one-story preparation and finishing module. Additionally, a smaller two-story
hexagonal module housing administrative and engineering support offices is located
in front of and connected to the assembly building.
 359
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

Receiving
and Engine assembly Final test
Shipping
storage
Spine

Machining Machining Machining Machining

Break
area
Spine Spine Spine

Figure 6.35 Block layout of the Volvo Skovdeverken factory.

The assembly operations are performed adjacent to the outer walls of the
three assembly modules. The assembly paths are indicated in Figure 6.37. From
Figure 6.38, notice the location of the material storage area in the center of the
plant. Lift trucks are used to store and retrieve materials in the storage racks, as
well as to transport materials between the storage area and pickup and deposit sta-
tions located on each floor. The lift truck operates on the first floor and lifts/lowers
materials to/from the second floor.
The Volvo Kalmar plant is recognized internationally for its pioneering work
on job enlargement and the team approach to automotive assembly. When the plan-
ning for the Kalmar plant began in the early 1970s, Pehr G. Gyllenhammar, manag-
ing director of the Volvo Group, gave the following general directive:

Final test area

Preparation and
Assembly module
finishing module

Assembly module Assembly module

Administrative
module

Figure 6.36 Volvo’s modular plant in Kalmar, Sweden.

 360
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

Figure 6.37 Assembly path at Volvo’s Kalmar plant. (Courtesy of Volvo.)

It has to be possible to create a working place which meets the need of the modern
human being for motivation and satisfaction in his daily work. It must be possible to
accomplish this objective without reducing efficiency [71].

The team concept was one of the basic objectives established at the beginning
of the facilities planning process. The organization was to be built around the team
concept. The team members were to collaborate on a common set of tasks and
work within an established production framework. They were to be allowed to
switch jobs among themselves, vary their work pace, carry joint responsibility for
quality, and have the possibility to influence their working environment. It was felt
by Volvo’s management that the larger number of work tasks, combined with team
membership, would provide greater meaning and satisfaction for each employee.
As a result of the planning objectives, each work team was provided its own
entrance, its own changing room and saunas, its own break area, and its own as-
sembly area. Each assembly area is approximately 10,000 ft2 in size and is viewed as
the team’s own small workshop.
Assembly is performed by 20 different teams. Each team completes one system
in the car, for example, the electrical system, instruments, and safety equipment. As-
sembly is carried out on special wire-guided, battery-powered, computer-controlled
carriers (automated guided-vehicle system, AGVS).
 361
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

Figure 6.38 Volvo’s Kalmar plant. (Courtesy of Volvo.)

Two assembly approaches are used. As illustrated in Figure 6.39, straight-line

assembly and dock assembly methods are used. With straight-line assembly, the
work to be performed by the team is divided among four or five workstations. The
workers work in pairs and follow a car from station to station carrying out the en-
tire work assignment belonging to their team. When a pair of workers complete
work on one car, they walk back to the beginning station in their area and repeat

Assembly stations

AGVS guide path

Incoming buffer Outgoing buffer

(a)

Assembly stations

AGVS guide path

Incoming Outgoing
buffer buffer
(b)

Figure 6.39 Assembly approaches used at Volvo’s Kalmar plant. (a) Straight-line assembly.
(b) Dock assembly.
 362
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

the process. Typically, the two-person team members trade work assignments to
provide additional variety to their work.
The dock assembly approach is used when the entire assembly task for an
area is carried out at one of the four assembly stations by a team of two or three
workers. With the dock assembly approach, the AGVS brings a car to an assembly
dock where the entire work cycle is performed. The content and quality of the work
performed is no different than that for the straight-line assembly approach.

6.10 DEVELOPING LAYOUT ALTERNATIVES

Because the final layout will result from the generation of alternative facility layouts,
it is important that the number, quality, and variety of alternative designs be as large
as feasible. If one agrees that facilities planning involves both art and science, then
there would surely be agreement that “the art” comes in generating layout alterna-
tives. It is at this stage that one’s creativity is truly tested.
We have previously emphasized the need to divorce one’s thinking from the
present method. The ideal system approach was intended to facilitate the creative
generation of alternatives. Aside from the algorithm-based procedures described in
the previous sections, there exist a number of aids to improve one’s ability to gen-
erate more and better design alternatives. The following are some suggested ap-
proaches that have proven to be beneficial:
1. Exert the necessary effort to generate and evaluate the alternatives. Do not
prematurely favor one solution; do not prematurely reject a solution either.
2. Set a time limit and use the available time wisely by assessing promising ver-
sus poor alternatives without getting bogged down in the details.
3. Seek many alternatives. In searching for good solutions to problems, Toyota
reportedly generates seven alternatives.
4. Establish a goal and layout performance metrics that the stakeholders agree on.
5. Make liberal use of the questioning attitude. The “5 Whys” used in lean man-
ufacturing can be a very powerful tool.
6. Don’t “fail to see the forest for the trees.”
7. Think big, then think little.
8. Don’t be conservative. Challenge legacy constraints and physical constraints
(such as existing walls within the building).
9. Refer to analogous problems of others.
10. Consult the literature (including the Web) and your peers in other organizations.
11. Use the brainstorming technique.
12. Divorce your thinking from the existing solution.
13. Involve operating people, management, experienced people, inexperienced
people, those who oppose change, and those who promote change.
14. Be aware of market trends and what the competition is doing.
15. Recognize your own limitations and strengths.
 363
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

16. Do your homework. Obtain reliable data/information and understand the

requirements.
17. Don’t overlook an improved version of the present method.
18. Think long range.
The number of possible facility layouts for any reasonably sized problem can
be quite large. Furthermore, the constraint specifications and the objective function
can often be “fuzzy.” For this reason, the design process does not ask for the selec-
tion of the best design from among all possible feasible designs. Rather, it asks for a
selection to be made from among several reasonable designs—it asks for the de-
signer to “satisfice” not “optimize.”
The problem of designing a facility layout can be distinguished from a number
of other problem-solving activities. As observed by Simon, “In ordinary language
. . . we apply the term ‘design’ only to problem solving that aims at synthesizing new
objects. If the problem is simply to choose among a given set of alternatives, e.g., to
choose the location or site for a plant, we do not usually call it a design problem,
even if the set of available alternatives is quite large, or possibly infinite” [60, p. 295].
Simon gives two reasons for calling the layout problem, but not a math pro-
gramming problem, a design problem. The negative reason he gives is that no sim-
ple finite algorithm exists for obtaining directly a solution to the layout problem.
The positive reason given by Simon is that the process used for attacking the layout
problem involves synthesizing the solution from intermediate, or component, deci-
sions that are “selective, cumulative, and tentative” [60].
Because of the immense number of possible layout designs available to the
facilities planner, the search process becomes a search over a space of design
components and partially completed designs, rather than a search over a space of
designs. In following the design process, very few complete designs are gener-
ated, compared, and evaluated. Instead, one generates design components, for
example, receiving and shipping component, assembly component, and manu-
facturing component. Even at the component level, the comparisons and evalua-
tions are typically performed at a macrodesign level. Quite often, only one
complete design is produced. The sequential process of generating and evaluat-
ing component designs is sufficiently effective as a filter that the complete design
is the final design.

6.11 SUMMARY
The objectives of this chapter were fourfold. First, we felt it important to cover the
“tried-and-true” material concerning the basic types of departments/layouts and al-
ternative layout planning procedures. We stressed that the department definitions
and types should be driven by the future state map. Second, we wanted to present
more formal layout algorithms and discuss their relative strengths and weaknesses.
Third, we wanted to emphasize the importance of planning for change. Change is
the only thing that is certain to occur in the future, and planning for change will
help guarantee that an organization is able to adapt and remain competitive. Fourth,
we wished to encourage you to “enter the world of the architect.”
 364
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

Aspects of form and style as they relate to facilities planning should not be dis-
missed out of hand. The architect typically approaches the layout planning problem
from a different perspective than does the facilities planning engineer or layout
analyst. It is important for the engineer to understand the architect’s perspective.
Furthermore, it has been our experience that the success of the layout analyst in
performing facilities planning frequently depends on his or her ability to work to-
gether with the architect.
Among the layout algorithms we covered in this chapter, CRAFT, MCRAFT,
and MULTIPLE use the discrete layout representation, and with the possible excep-
tion of MCRAFT, they place no restrictions on department shapes. Also, CRAFT and
MULTIPLE can capture the initial layout, the building shape, and fixed depart-
ments/obstacles with fairly high accuracy. However, with CRAFT and MULTIPLE, it
is difficult to generate departments with prescribed shapes (such as rectangular or
L-shaped departments). Although the spacefilling curve in MULTIPLE may be re-
vised to correct some department shapes, both algorithms are likely to generate
layouts that require considerable massaging.
BLOCPLAN, MIP, and LOGIC, on the other hand, use the continuous layout
representation, and they work only with rectangular department shapes. (LOGIC
may generate one or more nonrectangular departments if the building is nonrectan-
gular or if there are obstacles present.) Rectangular departments considerably in-
crease our control over department shapes. Also, in practice, a rectangular shape is
very likely to be acceptable (if not desirable) for many departments in a facility.
However, if fixed departments or obstacles are present, maintaining rectangular
departments may increase the cost of the layout. Furthermore, capturing the initial
layout, which may contain one or more nonrectangular departments, may not be
possible or straightforward (especially with BLOCPLAN and LOGIC).
We also demonstrated the fundamental “idea” behind each algorithm. For ex-
ample, MCRAFT relies on a sweep technique, while BLOCPLAN uses two or three
(horizontal) bands. LOGIC, on the other hand, divides the building progressively
into smaller portions by performing vertical and horizontal cuts and by assigning
one or more departments to each portion of the building. MULTIPLE uses a space-
filling curve (or a hand-generated curve), which “maps” the two-dimensional layout
into a single dimension (i.e., the layout vector). Last, MIP is based on mathematical
programming, and it uses 0/1 integer variables to ensure that departments do not
overlap.
As far as the objective function is concerned, all the algorithms we showed
use either the distance-based objective or the adjacency-based objective, or both. In
reality, almost any layout algorithm can be adapted to accept one objective function
or another. For example, the objective functions of CRAFT, LOGIC, and MULTIPLE
can be changed from distance based to adjacency based with no fundamental
changes to the algorithm itself. In fact, except for MIP, any one of the above algo-
rithms can be used to generate a layout and evaluate it with respect to a user-
specified objective function. One can also use an alternate objective function with
MIP; however, care must be exercised to maintain a “reasonable” objective function
(or at least a linear objective function) since the objective function, unlike the other
algorithms, is an inherent part of MIP.
For those readers who might, at least from a theoretical standpoint, be con-
cerned with layout optimality, we note that only two of the algorithms we presented
 365
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

can be used to identify the optimal layout: MIP will find the optimal layout only if all
the departments are required to be rectangular; MULTIPLE will find the optimal layout
only with respect to a particular spacefilling curve by enumerating over all possible de-
partment sequences in the layout vector. LOGIC may also find the “optimal” layout (by
enumerating over all possible cut-trees); however, if all the departments are rectangu-
lar, the cost of any optimal solution obtained by LOGIC will be greater than or equal
to the cost of an optimal solution obtained by MIP. (Why? Hint: consider the impact of
the “guillotine cuts” in LOGIC.)
There are a number of early layout algorithms that we did not describe in this
chapter. Such algorithms include ALDEP [57], COFAD [67], CORELAP [37], and
PLANET [11]. Since their introduction, these algorithms have served as the corner-
stones of facility layout algorithms, and they fueled interest in computer-based facil-
ity layout. We also did not present a number of other layout algorithms; see, for
example, DISCON [13], FLAC [56], and SHAPE [24], [68]. The interested reader may
refer to [14], [36], and [39] for a survey and discussion of single-floor and multi-floor
layout algorithms.
Some readers, from a theoretical or practical standpoint, might wonder which
layout algorithm is the best. From a theoretical standpoint, it is valid to compare cer-
tain layout algorithms. For example, since they use the same objective function and
the same representation, one may compare the performance of CRAFT, MCRAFT,
and MULTIPLE. In fact, in [7], numerical results are shown to compare CRAFT with
MULTIPLE. Likewise, as long as all the departments are rectangular, one may per-
form a meaningful comparison between BLOCPLAN, LOGIC, and MIP (assuming
that MIP is solved through a heuristic procedure such as the one shown in [46]). In
contrast, comparing, say, MULTIPLE with MIP requires a more careful approach; the
two algorithms use different representations, and the resulting department shapes
are likely to be different. Hence, any comparison must include the department
shapes. However, if the purpose of the comparison is to simply measure the layout
cost impact of allowing nonrectangular departments (at the risk of obtaining some
irregularly shaped departments), then it may make sense to compare just the layout
costs of MULTIPLE and MIP.
From a practical standpoint, we hope that the reader is now in a position to
appreciate that each layout algorithm has certain strengths and weaknesses. Partic-
ular constraints imposed by a problem (such as the number and location of fixed
departments and/or obstacles, department shapes, the building shape, moving into
a vacant building versus improving a given layout), coupled with our description of
each algorithm, should help the reader (or the layout planner) select not the “best”
algorithm but the “most appropriate” one. Although “dummy or negative flows” can
be used in a from-to chart, and some qualitative interactions among the depart-
ments may be captured in a relationship chart, we also need to remind the practi-
tioner that no computer-based layout algorithm will capture all the significant
aspects of a facility layout problem.
In fact, we do not know if computers will ever be able to fully capture and use
human experience and judgment, which play a critical role in almost any facility
layout problem. In that sense, a computer-based layout algorithm should not be
viewed as an “automated design tool” that can functionally replace the layout plan-
ner. To the contrary, we believe the (human) layout planner will continue to play a
key role in developing and evaluating the facility layout. The computer-based
 366
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

layout algorithms presented in this chapter (and other such algorithms in the litera-
ture) are intended as “design aids” that, when used properly, significantly enhance
the productivity of the layout planner. Therefore, for nontrivial problems, we be-
lieve that, given two layout planners with comparable experience and skills, the one
“armed” with an appropriate computer-based layout algorithm is far more likely to
generate a demonstrably “better” solution in a shorter time.

REFERENCES
1. Anjos, M. F., and Vannelli, A., “A New Mathematical-Programming Framework for Fa-
cility Layout Design,” INFORMS Journal on Computing, vol. 18, no. 1, pp. 111–118,
2006.
2. Apple, J. M., Plant Layout and Material Handling, 3rd ed., John Wiley, New York, 1977.
3. Armour, G. C., and Buffa, E. S., “A Heuristic Algorithm and Simulation Approach to Rel-
ative Location of Facilities,” Management Science, vol. 9, no. 2, pp. 294–309, 1963.
4. Bartholdi, J. J., and Platzman, L. K., “Design of Efficient Bin-Numbering Schemes for
Warehouses,” Material Flow, vol. 4, pp. 247–254, 1988.
5. Bartholdi, J. J., and Platzman, L. K., “Heuristics Based on Spacefilling Curves for
Combinatorial Problems in Euclidean Space,” Management Science, vol. 34, no. 3,
pp. 291–305, 1988.
6. Bozer, Y. A., and Meller, R. D., “A Reexamination of the Distance-Based Facility Layout
Problem,” IIE Transactions, vol. 29, no. 7, pp. 549–560, 1997.
7. Bozer, Y. A., Meller, R. D., and Erlebacher, S. J., “An Improvement-Type Layout Algo-
rithm for Single and Multiple Floor Facilities,” Management Science, vol. 40, no. 7,
pp. 918–932, 1994.
8. Brown, A. R., Optimum Packing and Depletion, Macdonald and Co., American Elsevier
Publishing Co., New York, 1971.
9. Buffa, E. S., Armour, G. C., and Vollman, T. E., “Allocating Facilities with CRAFT,” Har-
vard Business Review, vol. 42, no. 2, pp. 136–159, March/April 1964.
10. Davis, L. (ed.), Handbook of Genetic Algorithms, Van Nostrand Reinhold, New York,
1991.
11. Deisenroth, M. P., and Apple, J. M., “A Computerized Plant Layout Analysis and Evalu-
ation Technique (PLANET),” Technical Papers 1962, American Institute of Industrial
Engineers, Norcross, GA, 1972.
12. Donaghey, C. E., and Pire, V. F, “Solving the Facility Layout Problem with BLOCPLAN,”
Industrial Engineering Department, University of Houston, TX, 1990.
13. Drezner, Z., “DISCON: A New Method for the Layout Problem,” Operations Research,
vol. 20, pp. 1375–1384, 1980.
14. Drira, A., Pierreval, H., and Hajri-Gabouj, S., “Facility Layout Problems: A Survey,” Annual
Reviews in Control, vol. 31, no. 2, pp. 255–267, 2007.
15. Foulds, L. R., “Techniques for Facilities Layout: Deciding Which Pairs of Activities
Should Be Adjacent,” Management Science, vol. 29, no. 12, pp. 1414–1426, 1983.
16. Foulds, L. R., Gibbons, P. B., and Giffin, J. W., “Facilities Layout Adjacency Determina-
tion: An Experimental Comparison of Three Graph Theoretic Heuristics,” Operations
Research, vol. 33, no. 5, pp. 1091–1116, 1985.
17. Foulds, L. R., and Robinson, D. F., “Graph Theoretic Heuristics for Plant Layout Prob-
lem,” International Journal of Production Research, vol. 16, pp. 27–37, 1978.
18. Francis, R. L., McGinnis, L. F., and White, J. A., Facility Layout and Location: An Ana-
lytical Approach, Prentice Hall, Englewood Cliffs, NJ, 1992.
 367
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

19. Freeman, H., “Computer Processing of Line-Drawing Images,” Computing Surveys, vol. 6,
pp. 57–97, 1974.
20. Giffin, J. W., Foulds, L. R., and Cameron, D. C., “Drawing a Block Plan from a Relation-
ship Chart with Graph Theory and Microcomputer,” Computers and Industrial Engi-
neering, vol. 10, pp. 109–116, 1986.
21. Goldberg, D. E., Genetic Algorithms in Search, Optimization, and Machine Learning,
Addison-Wesley, Boston, 1989.
22. Harmon, R. L., and Peterson, L. D., Reinventing the Factory: Productivity Break-
throughs in Manufacturing Today, Free Press, New York, 1990.
23. Hassan, M. M. D., and Hogg, G. L., “On Constructing a Block Layout by Graph Theory,”
International Journal of Production Research, vol. 29, no. 6, pp. 1263–1278, 1991.
24. Hassan, M. M. D., Hogg, G. L., and Smith, D. R., “SHAPE: A Construction Algorithm for
Area Placement Evaluation,” International Journal of Production Research, vol. 24,
pp. 1283–1295, 1986.
25. Heragu, S. S., and Kusiak, A., “Efficient Models for the Facility Layout Problem,” Euro-
pean Journal of Operational Research, vol. 53, pp. 1–13, 1991.
26. Hobson, E. W., The Theory of Functions of a Real Variable and the Theory of Fourier’s
Series, Volume I, 3rd ed., Cambridge University Press and Harren Press, Washington,
DC, 1950.
27. Holland, J. H, Adaptation in Natural and Artificial Systems, University of Michigan
Press, Ann Arbor, MI, 1975.
28. Hosni, Y. A., Whitehouse, G. E., and Atkins, T. S., MICRO-CRAFT Program Documen-
tation, Institute of Industrial Engineers, Norcross, GA.
29. Immer, J. R., Layout Planning Techniques, McGraw-Hill, New York, 1950.
30. Jajodia, S., Minis, I., Harhalakis, G., and Proth, J., “CLASS: Computerized Layout Solu-
tions Using Simulated Annealing,” International Journal of Production Research, vol.
30, no. 1, pp. 95–108, 1992.
31. Johnson, D. S., Aragon, C. R., McGeoch, L. A., and Schevon, C., “Optimization by Sim-
ulated Annealing: An Experimental Evaluation; Part I, Graph Partitioning,” Operations
Research, vol. 37, pp. 865–892, 1989.
32. Johnson, R. V., “SPACECRAFT for Multi-Floor Layout Planning,” Management Science,
vol. 28, no. 4, pp. 407–417, 1982.
33. Kaku, K., Thompson, G. L., and Baybars, I., “A Heuristic Method for the Multi-Story
Layout Problem,” European Journal of Operational Research, vol. 37, pp. 384–397,
1988.
34. Kirkpatrick, S., Gelatt, C. D., and Vecchi, M. P., “Optimization by Simulated Annealing,”
Science, vol. 220, pp. 671–680, 1983.
35. Krejcirik, M., Computer Aided Plant Layout, Computer Aided Design, vol. 2, pp. 7–19,
1969.
36. Kusiak, A., and Heragu, S. S., “The Facility Layout Problem,” European Journal of
Operational Research, vol. 29, pp. 229–251, 1987.
37. Lee, R. C., and Moore, J. M., “CORELAP—Computerized Relationship Layout Planning,”
Journal of Industrial Engineering, vol. 18, no. 3, pp. 194–200, 1967.
38. Liggett, R. S., and Mitchell, W. J., “Optimal Space Planning in Practice,” Computer
Aided Design, vol. 13, pp. 277–288, 1981.
39. Meller, R. D., “Layout Algorithms for Single and Multiple Floor Facilities,” Ph.D. Disser-
tation, Department of Industrial and Operations Engineering, University of Michigan,
1992.
40. Meller, R. D., personal communication, 1993.
41. Meller, R. D., and Bozer, Y. A., “A New Simulated Annealing Algorithm for the Facil-
ity Layout Problem,” International Journal of Production Research, vol. 34, no. 6,
pp. 1675–1692, 1996.
 368
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

42. Meller, R. D., and Bozer, Y. A., “Alternative Approaches to Solve the Multi-Floor Facil-
ity Layout Problem,” Journal of Manufacturing Systems, vol. 16, no. 3, pp. 192–203,
1997.
43. Meller, R. D., Narayanan, V., and Vance, P. H., “Optimal Facility Layout Design,” Oper-
ations Research Letters, vol. 23, no. 3–5, pp. 117–127, 1998.
44. Metropolis, N., Rosenbluth, A. W., Rosenbluth, M. N., and Teller, A. H., “Equation of
State Calculation by Fast Computing Machines,” Journal of Chemical Physics, vol. 21,
pp. 1087–1092, 1953.
45. Montreuil, B., “A Modeling Framework for Integrating Layout Design and Flow Net-
work Design,” Proceedings of the Material Handling Research Colloquium, pp. 43–58,
Hebron, KY, 1990.
46. Montreuil, B., and Ratliff, H. D., “Utilizing Cut Trees as Design Skeletons for Facility
Layout,” IIE Transactions, vol. 21, no. 2, pp. 136–143, 1989.
47. Montreuil, B., Ratliff, H. D., and Goetschalckx, M., “Matching Based Interactive Facility
Layout,” IIE Transactions, vol. 19, no. 3, pp. 271–279, 1987.
48. Murty, K., Linear and Combinatorial Programming, Wiley, New York, 1976.
49. Muther, R., Systematic Layout Planning, 2nd ed., Cahners Books, Boston, 1973.
50. Nugent, C. E., Vollman, R. E., and Ruml, J., “An Experimental Comparison of Tech-
niques for the Assignment of Facilities to Locations,” Operations Research, vol. 16,
pp. 150–173, 1968.
51. Pire, V. F., “Automated Multistory Layout System,” Master’s Thesis, Industrial Engineer-
ing Department, University of Houston, TX, 1987.
52. Propst, R. L., The Action Factory System: An Integrated Facility for Electronics Manufac-
turing, Herman Miller Corp., Zeeland, MI, 1981.
53. Reed, R., Jr., Plant Layout: Factors, Principles and Techniques, Richard D. Irwin, IL, 1961.
54. Rother, M., and Shook, J., Learning to See: Value Stream Mapping to Add Value and
Eliminate MUDA, Lean Enterprise Institute, Cambridge, MA, 1999.
55. Schonberger, R. J., World Class Manufacturing, The Lessons of Simplicity Applied, Free
Press, New York, 1986.
56. Scriabin, M., and Vergin, R. C., “A Cluster-Analytic Approach to Facility Layout,” Man-
agement Science, vol. 31, no. 1, pp. 33–49, 1985.
57. Seehof, J. M., and Evans, W. O., “Automated Layout Design Program,” Journal of In-
dustrial Engineering, vol. 18, pp. 690–695, 1967.
58. Seppannen, J., and Moore, J. M., “Facilities Planning with Graph Theory,” Management
Science, vol. 17, no. 4, pp. 242–253, December 1970.
59. Sherali, H. D., Fraticelli, B. M. P., and Meller, R. D., “Enhanced Model Formulations for
Optimal Facility Layout,” Operations Research, vol. 51, no. 4, pp. 629–644, 2003.
60. Simon, H. A., “Style in Design,” in C. M. Eastman (ed.), Spatial Synthesis in Computer-
Aided Building Design, Wiley, New York, 1975.
61. Skinner, W., “The Focused Factory,” Harvard Business Review, pp. 113–121, May–June
1974.
62. Tam, K. Y, “A Simulated Annealing Algorithm for Allocating Space to Manufacturing
Cells,” International Journal of Production Research, vol. 30, pp. 63–87, 1991.
63. Tanaka, H., and Yoshimoto, K., “Genetic Algorithm Applied to the Facility Layout Prob-
lem,” Department of Industrial Engineering and Management, Waseda University,
Tokyo, 1993.
64. Tanchoco, J. M. A. (ed.), Material Flow Systems in Manufacturing, Chapman and Hall,
London, 1994.
65. Tate, D. M., and Smith, A. E., “A Genetic Approach to the Quadratic Assignment Problem,”
Computers and Operations Research, vol. 22, no. 1, pp. 73–83, 1995.
66. Tate, D. M., and Smith, A. E., “Unequal-Area Facility Layout by Genetic Search,” IIE
Transactions, vol. 27, no. 4, pp. 465–472, 1995.
 369
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

67. Tompkins, J. A., and Reed, R., Jr., “An Applied Model for the Facilities Design Problem,”
International Journal of Production Research, vol. 14, no. 5, pp. 583–595, 1976.
68. Tretheway, S. J., and Foote, B. L., “Automatic Computation and Drawing of Facility
Layouts with Logical Aisle Structures,” International Journal of Production Research,
vol. 32, no. 7, pp. 1545–1555, 1994.
69. “Vertical Conveyors Increase Plant’s Efficiency 33%,” Modern Materials Handling,
vol. 40, Casebook Directory, p. 113, 1985.
70. “Vertical Reciprocating Conveyors: Flexible Handling Devices,” Material Handling En-
gineering, vol. 45, no. 9, pp. 81–85, 1990.
71. Volvo: The Engine Factory That Is Different, a Volvo publication describing the Volvo
Skovde plant, Gothenberg, Sweden.
72. White, J. A., “Layout: The Chicken or the Egg?,” Modern Materials Handling, vol. 35,
no. 9, p. 39, September 1980.
73. Wilhelm, M. R., and Ward, T. L., “Solving Quadratic Assignment Problems by Simulated
Annealing,” IIE Transactions, vol. 19, pp. 107–119, 1987.

PROBLEMS
SECTION 6.1–6.2
6.1 What are some of the important factors that should be taken into consideration when a
layout is being designed?
6.2 In a manufacturing environment, what is the impact of the material handling decisions
on the effectiveness of a facility layout?
6.3 What kind of manufacturing environments are the following types of layout designs
best suited for?
a. Fixed product layout
b. Product layout
c. Group layout
d. Process layout
6.4 Compare the primary layout design objectives for the following situations:
a. Soda bottler
b. Printing shop
c. Meat-processing plant
d. Furniture manufacturing plant
e. Computer chip maker
f. Shipyard
g. Refinery plant
h. College campus

SECTION 6.3
6.5 What are the basic differences between construction-type and improvement-type
layout algorithms?
6.6 Contrast and compare the plant layout procedures proposed by Apple, Reed, and
Muther.
6.7 Four departments are to be located in a building of 600  1000. The expected per-
sonnel traffic flows and area requirements for the departments are shown in the tables
below. Develop a block layout using SLP.
 370
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

Dept. A B C D
A 0 250 25 240
B 125 0 400 335
C 100 0 0 225
D 125 285 175 0

Department Department Dimension

A 200  200
B 400  400
C 600  600
D 200  200

6.8 XYZ Inc. has a facility with six departments (A, B, C, D, E, and F). A summary of the
processing sequence for 10 products and the weekly production forecasts for the prod-
ucts are given in the tables below.
a. Develop the from-to chart based on the expected weekly production.
b. Develop a block layout using SLP.

Product Processing Sequence Weekly Production

1 ABCDEF 960
2 ABCBEDCF 1200
3 ABCDEF 720
4 ABCEBCF 2400
5 ACEF 1800
6 ABCDEF 480
7 ABDECBF 2400
8 ABDECBF 3000
9 ABCDF 960
10 ABDEF 1200

Dept. Dimension
A 40  40
B 45  45
C 30  30
D 50  50
E 60  60
F 50  50

6.9 A toy manufacturing company makes 10 different types of products. There are 15
equal-sized departments involved. Given the following product routings and produc-
tion forecasts,
a. Construct a from-to chart for the facility.
b. Develop a block layout using SLP.
 371
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

Product Processing Sequence Weekly Production

1 ABCDBEFCDH 500
2 MGNONO 350
3 HLHK 150
4 CFEDH 200
5 NON 100
6 IJHKL 150
7 GNO 200
8 ACFBEDHD 440
9 GMN 280
10 IHJ 250

6.10 Shown below is the activity relationship chart along with the space requirements for
each of the six cells in a small auto parts manufacturing facility.
a. Construct relationship and space relationship diagrams.
b. Design the corresponding block layout using SLP.

6.11 An activity relationship chart is shown below for the American Mailbox Company. Con-
struct a relationship diagram for the manufacturing facility. Given the space require-
ments (in ft2), construct a block layout using SLP.

SECTION 6.4
6.12 Suppose five departments labeled A through E are located as shown in the layout
below. Given the corresponding flow-between chart, compute the efficiency rating for
the layout.
 372
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

A B C D E
A — 5 0 4 3
B — 6 1 2
A B C — 6 0
C D — 3
D E E —

6.13 In an assembly plant, material handling between departments is performed using a uni-
directional closed-loop conveyor. The figure below shows the layout for the modular

facility, which consists of three equal-sized assembly modules (A, B, and C), one admin-
istrative module (D), and one warehouse module (E). P/D points for each module are
also shown in the figure. The administrative and warehouse activities are not to be
moved; however, assembly areas A, B, C can be relocated. The distance between P/D
points and the number of pallet loads moved between departments are given below.

Distance between P/Ds Pallet Flow per Day

From To Distance From/To A B C D E
P/D 1 P/D 2 60 A 0 0 5 0 30
P/D 2 P/D 3 90 B 10 0 25 0 0
P/D 3 P/D 4 30 C 25 5 0 0 0
P/D 4 P/D 5 90 D 0 0 0 0 0
P/D 5 P/D 1 60 E 5 20 5 0 0

Using the pairwise exchange method, determine new locations for assembly modules
A, B, and C that minimize the sum of the products of pallet flows and conveyor travel
distances.
6.14 Four equal-sized machines are served by an automated guided vehicle (AGV) on a
linear bidirectional track, as shown in the figure below. Each machine block is 30 
30. The product routine information and required production rate are given in the table
below. Determine a layout arrangement based on the pairwise exchange method. As-
sume that the pickup/delivery stations are located at the midpoint of the machine edge
along the AGV track.
 373
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

Product Processing Sequences Weekly Production

1 BDCAC 300 units
2 BDAC 700 units
3 DBDCAC 900 units
4 ABCA 200 units

6.15 Using the data from Problem 6.14 but assuming that the locations of the P/D points for
machines A and B are 5 feet from the lower righthand corner of each machine and the
P/D points for machines C and D are 10 feet from the lower-lefthand corner of ma-
chines C and D, develop an improved layout using the pairwise exchange method.
6.16 A mobile robot is serving two cells located at either sides of the AGV track, as shown
by the figure below. There are three machines placed in each cell. Given the from-to
chart in the table below, find the best machine arrangements for both cells. Rearrange-
ment is limited only to machines within each cell. Assume that the P/D point of each
machine is located at the midpoint of the machine edge along the AGV track.

M/C A B C D E F
A — 10 50 30 0 60
B 5 — 45 40 30 0
C 40 30 — 35 5 20
D 40 25 50 — 40 50
E 0 55 40 50 — 0
F 20 0 60 20 10 —

M/C Distance M/C Distance

A-B 30 D-E 30
A-C 60 D-F 60
B-C 30 E-F 30

6.17 Five machines located in a manufacturing cell are arranged in a “U” configuration as
shown in the layout below. The material handling system employed is a bidirectional
conveyor system. Determine the best machine arrangement given the product routing
information and production rates in the table.
 374
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

Product Machine Sequence Prod. Rate

1 B-E-A-C 100
2 C-E-D 200
3 B-C-E-A-D 500
4 A-C-E-B 150
5 B-C-A 200

M/C Distance M/C Distance

A-B 20 B-D 100
A-C 70 B-E 120
A-D 120 C-D 50
A-E 140 C-E 70
B-C 50 D-E 20

6.18 The ABC Cooling and Heating Company manufactures several different types of air con-
ditioners. Five departments are involved in the processing required for the products. A
summary of the processing sequences required for the five major products and the weekly
production volumes for the products are shown in the tables below along with the depart-
ment area. Based on the graph-based construction method, develop a block layout.

Product Process Sequence Weekly Production

1 ABC 150
2 ABED 200
3 ACE 50
4 ACBE 200
5 ADE 250

Department Area (ft2)

A 1500
B 1500
C 1000
D 2000
E 2000
 375
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

6.19 The activity relationship chart for Walter’s machine shop is shown in the figure below.
The space requirements are in square feet. Construct the relationship diagram and de-
velop a block layout using the graph-based method.

Dept. A B C D E F G H
A — 302 0 0 0 66 0 68
B 0 — 504 20 136 154 56 40
C 0 0 — 76 352 0 122 94
D 0 0 0 — 0 0 180 8
E 0 0 0 0 — 122 0 282
F 0 0 0 0 0 — 188 24
G 0 0 0 0 0 0 — 296
H 0 0 0 0 0 0 0 —

6.20 The from-to material flow matrix for an eight-department facility is given in the table
below. Construct a relationship diagram based on the material flow matrix, and con-
struct a block layout using the graph-based method.

Dept. Area Required (ft2)

A 2800
B 2100
C 2600
D 400
E 600
F 400
G 2300
H 1800

6.21 Consider the layout of five equal-sized departments. The material flow matrix is given
in the figure below.
a. Develop the final adjacency graph using the graph-based procedure.
b. Develop a block layout based on the final adjacency graph obtained in part a.
 376
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

A B C D E
A — 0 5 25 15
B 0 — 20 30 25
C 0 25 — 40 30
D 30 5 20 — 0
E 20 30 5 10 —

6.22 The material flow matrix for 10 departments is given below.

A B C D E F G H I J
A — 0 12 0 132 16 0 220 20 24
B 0 — 176 0 216 0 144 128 0 0
C 0 0 — 0 0 184 0 0 28 0
D 212 136 240 — 36 0 236 0 164 0
E 0 0 140 0 — 0 192 0 0 160
F 0 180 0 188 108 — 248 228 0 0
G 172 0 156 0 0 0 — 112 224 152
H 0 0 32 40 204 0 0 — 0 0
I 0 168 0 0 104 156 0 148 — 200
J 0 124 196 120 0 116 0 108 0 —

The area requirements are

Dept. Area (ft2)

A 400
B 1000
C 2600
D 400
E 2400
F 1000
G 3600
H 1200
I 400
J 2400

a. Determine a final adjacency graph using the graph-based procedure.

b. Construct a block layout based on the adjacency graph in part a.
6.23 Suppose the efficiency rating is designated by E, and the cost computed by CRAFT is
designated by K. Further suppose that four departments (labeled A, B, C, and D) are
given for a layout problem. Each department is assumed to be of equal area, and each
department is represented by a unit square. The unit cost used in computing K is
assumed to be equal to 1.0 for all department pairs. (That is, cij  1.0 for all i, j.) The
flow between the above departments is given as follows:
 377
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

A B C D
A — 10 10 0
B — 0 4
C — 4
D —

a. Compute the values of E and K for the following layout:

A C

B D

b. Compute the values of E and K for the following layout.

B
A D
C

c. Compute the values of E and K for the following layout.

A B
C D

d. What can you say regarding the consistency of the two measures. That is, if a layout
is “good” when measured by its E value, would it still be “good” when it is mea-
sured by its K value or vice versa? Justify your answer.
6.24 Consider four departments labeled A, B, C, and D. Each department is represented by
a 1  1 square. The following data are given:

Flow-Between Matrix Unit Cost Matrix

A B C D A B C D
A — 6 0 3 A — 2 0 3
Initial Layout B — 5 0 B 2 — 1 0
A B C — 0 C 0 1 — 0
C D D — D 3 0 0 —

The location of department A is fixed. Answer the following questions using CRAFT
with two-way exchanges only.
a. List all the department pairs that CRAFT would consider exchanging. (Do not com-
pute their associated cost.)
b. Compute the actual cost of exchanging departments C and D.
c. Given that department A is fixed and that each department must remain as a 1  1
square, is the layout obtained by exchanging departments C and D optimum? Why
or why not? (Hint: Examine the properties of the resulting layout and consider the
objective function of CRAFT.)
 378
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

6.25 The following layout is an illegal CRAFT layout. Nevertheless, given that the volume of
flow from A to B is 4, A to C is 3, and B to C is 9, and that all move costs are 1, what
is the layout cost?

C C C C C C
C B B B B C
C B A A B C
C B B B B C
C C C C C C

6.26 When CRAFT evaluates the exchange of departments, instead of actually exchanging
the departments, it only exchanges the centroids of departments.
a. What is the impact of this method of exchanging if all departments are the same
size?
b. Given the following from-to chart and scaled layout (each square is 1  1), what
does the evaluation of the exchange of departments B and C indicate should be
saved over the existing layout, and what is actually saved once this exchange is
made?

To
From A B C
A — 10 6 A A A C B B B B
B 2 — 7 A A A C B B B B
C — A A A C B B B B
From-To Chart Initial Layout

6.27 Explain the steps CRAFT would take with the following problem and determine the
final layout. Only two-way exchanges are to be considered.

To
From A B C D E
A — 3 2 1 A A A B B B
B — 1 3 A A A C C C
C 1 — 4 A A A C C C
D — D D D E E E
E — D D D E E E
From-To Chart Initial Layout

6.28 A manufacturing concern has five departments (labeled A through E) located in a rec-
tangular building as shown below:
 379
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

Suppose the flow data, the unit cost data, and the distance matrix are given as follows:

Flow-Between Matrix Unit Cost ($/Unit Dist.) Distance Matrix

A B C D E A B C D E A B C D E
A — 0 5 0 5 A — 0 1 0 1 A — 6 20 18 11
B — 6 2 0 B — 1 4 0 B — 12 22 15
C — 3 0 C — 3 0 C — 10 8
D — 7 D — 1 D — 7
E — E — E —

a. Using the CRAFT two-way exchange procedure, indicate all the department pairs
CRAFT would consider exchanging in the above layout.
b. Compute the estimated cost of exchanging departments A and E.
6.29 Suppose the following layout is provided as the initial layout to CRAFT. The flow-between
matrix and the distance matrix are given as follows. (All the cij values are equal to 1.0.)

Flow-Between Matrix Distance Matrix

A B C D E F A B C D E F
A — 0 8 0 4 0 A — 30 25 55 50 80
B — 0 5 0 2 B — 45 25 60 50
C — 0 1 0 C — 30 25 55
D — 6 0 D — 45 25
E — 4 E — 30
F — F —

a. Given the above data and initial layout, which department pairs will not be consid-
ered for exchange.
b. Compute the cost of the initial layout.
c. Compute the estimated layout cost assuming that departments E and F are exchanged.
d. In general, when the same data and initial layout are supplied to CRAFT and MUL-
TIPLE, why would one expect MULTIPLE to often (but not always) outperform
CRAFT? Also, what type of data and/or initial layout would allow CRAFT to consis-
tently generate layout costs that are comparable to those obtained from MULTIPLE?
6.30 Answer the following questions for CRAFT.
a. State two principal weaknesses and two principal strengths of CRAFT.
b. CRAFT uses the estimated cost in evaluating the potential impact of exchanging two
or three department locations. Suppose we modify the original computer code to
 380
Part Two DEVELOPING ALTERNATIVES: CONCEPTS AND TECHNIQUES

obtain a new code, say, NEWCRAFT, where we always use the actual cost in evalu-
ating the potential impact of exchanging two or three departments. Assuming that
we start with the same initial layout and that we use the same exchange option
(such as two-way exchanges only, or any other exchange option), the cost of the fi-
nal layout obtained from NEWCRAFT will not necessarily be less than the cost of the
final layout obtained from CRAFT. True or false? Why?
6.31 Using BLOCPLAN’s procedure, convert the following from-to chart to a relationship chart.

To
From A B C D E F G H
A — 8 3 6
B 1 — 5
C — 4
D 9 — 18
E 4 1 —
F 4 4 —
G 2 — 20
H 7 —

6.32 Consider BLOCPLAN. Suppose the following REL-chart and layout are given for a five
department problem. (It is assumed that each grid in the layout represents a unit
square.) Further suppose that the following scoring vector is being used: A  10, E 
5, I  2, O  1, U  0, and X  10.

1 2 3 4 5
1 — A U E U
2 — U U I
3 — U X
4 — A
5 —

a. Compute the “efficiency rating.”

b. Compute the REL-DIST score.
c. In improving a layout, BLOCPLAN can exchange only those departments that are
either adjacent or equal in area. True or false? Why? (If true, then explain why there
is such a limitation. If false, then explain how two departments that do not meet the
above constraint are exchanged.)
d. In constructing and improving a layout, BLOCPLAN maintains rectangular depart-
ment shapes. Discuss the advantages and limitations of maintaining such depart-
ment shapes in facility layout.
6.33 Use the MIP model to obtain an optimal layout with the data given for Problem 6.28.
(You may assume that each grid measures 20  20.)
6.34 Use the MIP model to obtain an optimal layout with the data given for Problem 6.29.
(You may assume that each grid measures 10  10.)
 381
6 LAYOUT PLANNING MODELS AND DESIGN ALGORITHMS

6.35 Consider the layout shown in Figure 6.26d. Use LOGIC and the cut-tree shown in
Figure 6.27 to exchange departments B and F.
6.36 Re-solve Problem 6.35 by exchanging departments D and H in the layout shown in
Figure 6.26d.
6.37 Re-solve Problem 6.35 by exchanging departments G and H in the layout shown in
Figure 6.29.
6.38 Answer the following questions for MULTIPLE.
a. Consider four departments (labeled A through D) with the following area requirements:
A  7 grids, B  3 grids, C  4 grids, and D  2 grids. Using the spacefilling curve
shown below, show the layout that would be obtained from the sequence A-B-C-D.

b. Using the data given for part a and the spacefilling curve shown above, show the
layout that would be obtained by exchanging departments B and D.
c. Discuss the advantages and limitations of using spacefilling curves in the manner
they have been used in MULTIPLE. Your discussion should include the treatment of
fixed departments, dummy departments, unusable floor space (i.e., obstacles), and
extra (i.e., empty) floor space.
6.39 Consider the initial layout and flow/cost data given for Problem 6.28.
a. Using the following conforming curve and MULTIPLE, improve the initial layout via
two-way department exchanges.

b. In general, what is the disadvantage of using conforming curves?

6.40 Consider the initial layout and flow data given for Problem 6.29. Use MULTIPLE and
the following conforming curve to obtain a two-opt layout.