You are on page 1of 12

Optimal space planning in

practice
Robin S Liggett and William J Mitchell

known to differ from an optimal arrangement with


The Computer Aided Design Group's Space Planning respect to some criterion, together with a knowledge
System, a general facility for optimal spatial arrangement of the costs of moving activities from one location to
of activities in buildings, is described. Casestudies of another, what activity moves are worth making, consider-
application in practice are presented and discussed, and ing the tradeoff between move costs and costs associated
the influence of practical experience with real spatial with an arrangement criterion? For example, as project
arrangement problems on the design of the software is groups increase or decrease in size, how should
described. employees be located within an office such that a
criterion of group contiguity is maintained with a
minimum number of work-space moves?
Algorithms to aid in the solution of a large class of problems
known as spatial allocation problems were first developed Solution of such problems can be handled as a single-stage
approximately 20 years ago. These algorithms have since or multi-stage process (Figure I). The problem of assigning
formed the basis of a number of computer programs activities to floors in a multi-storey office building is an
designed specifically for architectural applications, provid- example of a single-stage stacking plan problem. More
ing space planners with tools for producing stacking plans complex problems can be approached in multiple stages.
and schematic floor plan layouts for buildings. The objec- For example, activities might first be assigned to buildings,
tive of such programs is to locate a set of activities within then within each building activities might be assigned to
a facility (eg employees within an office space) such that floors, then within each floor activities might be assigned
operating efficiency is maximized. Operating efficiency to zones, then within each zone activities might be assigned
might be defined in terms of contiguity of working groups; to specific locations. The Space Planning System allows
ease of communication between employees; level of fixed either a single-stageor multi-stage approach to be taken,
operating expenses; amount of wasted and/or unused as appropriate.
space; or appropriate accommodation of growth, shrinkage The Space Planning System treats all such problems as
and change in spatial requirements over time. specialized forms of the general quadratic assignment
This paper describes the principles and applications of problem, which is concerned with finding optimal locations
the Computer Aided Design Group's Space Planning for a set of interrelated objects. It was first formulated in
System, a software system which produces high quality 1957 by Koopmans and Beckmann I for a plant location
solutions to both small and large space planning problems problem, where the objective was to minimize the cost
concerned with issuesof operating efficiency. of transporting goods among the plants.
Three different types of problems can be addressed In space planning versions of the quadratic assignment
with the system: problem, as dealt with by the Space Planning System,
there are three major components of cost:
• The stacking or zone plan optimization problem. Given
some division of a total facility into parts (for example • The fixed costs of assigning particular activities to
buildings, zones or floors), and a set of activities to be particular locations, for example the rent paid for
located within the facility, what is the best (optimum) the space that houses an activity.
way of assigning activities to parts of the facility? This • The interactive costs or communication costs between
problem most frequently arises with respect to multi- activities, for example the cost of the time spent by a
storey office buildings where activities must be assigned secretary in walking back and forth to a distant copying
to different floors, and thus it is generally known as the machine.
stacking plan problem. • The move cost of shifting activities from one place to
• The block plan optimization problem. Given a floor another in order to satisfy space needs, or to achieve
plan for a building or some part of a building, and a set a more efficient arrangement.
of activities to be located on that floor, what is the best An implicit enumeration approach to solution of quadratic
spatial arrangement? assignment problems, developed by Graves and Whinston2,
• The move optimization plan. Given an existing assign- forms the basis of the Space Planning System's solution
ment of activities to locations within a facility that is generation routines. This approach is very powerful
because it allows the same solution procedure to be used
Mitchell, Readerand Hamer: the Computer Aided DesignGroup, to assign activities at each stage of a multi-stage problem.
2407 Main Street, SantaMonica,CA 90405, USA A number of special features have been incorporated in

volume 13 number 5 september 1 9 8 1 0010-4485/81/050277-I 2 $02.00 © 1981 IPC Business Press :277
Stage I p= where p (&) =/k
Assign activities to buildings \h,/2, ,]m/ c/V
The quadratic assignment problem consists of finding, from
among the set of possible alternative solutions $, a map p
that optimizes the value of an objective function.
Let
F = lfi/} be an m xn matrix
:::::::. Q= [qij] b e a n m x m matrix, and
I ...... I
Stage 2 ::::::::. C = [c/j] b e a n n x n matrix
Assign activities to floors within I I
a single building where
I::::::: I
1....... I
fij is the fixed cost of assigning element iEM to
elementjeN,
qij is some measure of interaction between elements
i, jeM, and
cij is some measure of distance between elements i, jeN.
A general mathematical statement of the problem is
Stage 5
Assign activities to zones within a Min ~fip(i) + ~ ~qi/cp(i)p(/)
single floor
pe5 i i j

F o r m u l a t i o n of space planning p r o b l e m s
Formulation of a space planning problem as a quadratic
assignment problem involves consideration of the following
questions:
• How can the problem be expressed as a one-to-one
Sb0ge 4
assignment problem?
Assign activities to specific • How should zones and activities be defined?
locations within o zone • How should the cost function be specified?
• What type of travel time data should be used?
• How should interactions between activities be expressed?
• How should fixed costs be expressed?
Figure 1. Multi-stage space planning system These questions are considered, in turn, in the following
subsections. The discussion is illustrated by example
stacking plan and block plan problems.
the algorithm to deal with issues relevant to the space
planning problem that are not handled by the general Obtaining a one-to-one assignment problem
quadratic assignment formulation. The general quadratic assignment model, as specified
This paper discusses how formulations of space planning formally above, assumes a one-to-one assignment of
problems that are sufficiently realistic to reflect practical activities to locations. The optimal solution is one which
needs can be achieved, and how such realistic formulations minimizes an additive function of the fixed costs due to
complicate the solution synthesis process. Detailed examples the assignment of individual activities to individual
of applications of the Space Planning System to real projects locations (move costs can be incorporated as a special type
are presented, and it is shown how features of the system of fixed cost) and also of the interactive costs which are
have evolved from experience gained in practical dependent on the relative locations assigned to pairs of
applications. activities.
Space planning problems are more complex than tradi-
tional quadratic assignment problems due to the effects
PROBLEM FORMULATION of imposing area requirements. The areas required by
General quadratic assignment problem activities are not necessarily equal, and so it is not feasible
in general to match activities and locations on a one-to-one
Consider the quadratic assignment problem in terms of a basis. As a result, it is necessary to regard each space or
mapping of a set M = { 1,2 . . . . . m } into a set activity as composed of some number of equal-sized
N = { 1 , 2 , . . . ,n } whereto ~<n. Each element of M is modules according to required floor area. Correspondingly,
assigned to a distinct element of N. Thus the mapping is a each facility part (eg building, zone, or floor) is partitioned
one-to-one assignment of elements of one set to elements into location modules equal in size to the activity modules.
of another. A facilities location problem, for example, The problem is then one of assigning activity modules to
can be thought of as a mapping of a set of facilities into a location modules.
set of locations where each facility is assigned a distinct
location. Definition of zones and activities in stacking plan problems
Let S be the set of all possible mappings of the set M Figure 2 shows data defining zones and activities for a
into the set N. A particular map peS can be represented stacking plan problem for a large bank building. In this
in the form case zones represent floors of a multi-storey building, and

278 computer-aided design


the problem is one of assigning various activities to these of area modules. Zone definition must be given careful
floors. The user provides data describing each floor and thought. Architecturally meaningful zoning will result in
its area (in square feet or metres) and the size of the generation of better results at lower costs than arbitrary
desired area module. The program then determines the patterns of zoning. The maximum zone size for efficient
number of location modules allocated to each zone. operation is about 100 modules.
Similar input is given for each activity (see Figure 26). Preassignment of activities at the block plan level can
In addition, selected activities can be preassigned to also be handled by perimeter definition of a specific shape.
specific zones (floors) to reflect earlier design decisions In the example in Figure 3, Activity 1 has been preassigned
or known requirements. In this example a number of to the northeast corner of the plan.
employee related activities, such as food facilities and the
nursing station, have been preassigned to floor 11, which Specification of the cost function
is the elevator transfer floor. Preassigned activities are The objective, in generating an arrangement using the
never shifted by the computer program. Space Planning System, is to produce an assignment of
activities to zones (eg to floors or to specific locations on
a floor) that minimizes an overall cost measure, subject
Definition of zones and activities for block plan problems
At the block.plan level (layout of a single floor), a two-step to meeting specified space needs requirements. The cost
solution strategy normally is required due to computation measure takes into account both fixed costs and inter-
and computer memory requirements. First activities are active costs. Interactive costs are calculated as the product
assigned to subdivisions (zones) of the floor plan, and of some measure of interaction between pairs of activities
then to specific location modules within these subdivisions. and the distance or travel time between the locations to
which they have been assigned. This implies that activities
Figure 3 illustrates zone definition for the layout of a
which are closely interrelated will normally be located
single floor of a department store. In this case zones can
be specified by defining their physical perimeters in terms close together, while activities which are not closely inter-
related will tend to be separated.
ZORE DEFINITION
To provide a basis for computation of this objective,
~RKi # OF it is necessary to input travel time, interaction, and fixed
LO# {S~ FEET ) ~ESCRLPTION MODULES
....................................................................
~ 11093 FLOOR |2,?LAZA 115
cost data.
19137 FLCGN 13,HEZZANINE 198
3 3 17729 RLOOR '~ 183
~ 18233 FLCON IS 186
18230 FLOOR 16
18233 FLOOR 17
180
168 Travel time data
18230 FLO0~ 18 168
8
9
6
9
18233 FLCCR #9
18230 FLOOR 110
188
188
A matrix representing distance or travel time between
13 10
11 11
17792 FLOCR ~II,TNANS~ER
1E1~0 FLOOR #12
169
187 zones must be input by the user, except in the case where
12 12 1759~ FLCC6 #1~ 182
13 13
lq Iq
18212 FLO~B #15
18661 FLCLR 116
188
1$3
zone perimeters have been specified, as in the example
15 15 FLOOR t 1 7
16 16
16661
18661 FLGCR 119
193
193 block plan problem. In this latter case, distances can be
17 17 20101 FLOOR #51 207
18 18
19 19
2015~
22656
FLCCR #52
FLOOR #53
2OE
237
calculated automatically by the program as either
20 23 19~21 FLCCR 159 230
straight-line or city-block distances between zone centroids
TOTAL: 367703 37~6
(a multiplication factor can be input to translate distance

RCTIVITT DB~IBITION
AREA | OF P R E A S S I G N E D TO ZONE: | OF
IDR (5~ F R E T ) D E S C R I P T I C R NODULES IDR DESCRIPTION ROUBLES
..................... .-:-. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1 1 15127 CORRORATE BARKING R10 156
2 2 13197 INIERR&TIORAL,OPERATIOR$ glO 136
3 8799 IRTERRATIGRAL.BARNETIRG E20 91
N ;1993 TRDS~,ADBIRISTRATIOR C10 227
5 28897 T R R S T . C E R T . ACCTG. C20 298
6 6 1686 ERPLOTER T R U S T C30 17
7 7 5705 R~ALESTRTErzi. IRSTIT. 010 59
8 8 7653 CONSTRUCTION F2NANCE B20 79
9 9 2073 SEAL E S T A T E & P R R . , D E S I D ° ZlO 21
10 10 3551 REAL E S T A T E A P P R o , I R C O M E R20 37
11 I1 3131 AUDITING E30 32
12 12 10603 COMMMRCIAL MORTGAGE PIO 109
18 13 12633 DISTRICT 6 F20 130
lq 1~ 2075 8RANCH PLARNING FRO 21
15 15 5091 BRANCH A D M i R I S T RATION FRO 53
16 16 11005 L I A B I L I T T CARD F50 11q
17 17 1790 ~SCRON F60 18
18 18 3483 LOAH-ADMINIE~RAIION F70 35
19 19 3836 INDIRECT L O A R - A D M I R . E80 qO
~0 20 2717 ADEIRISTRATION F90 28
21 21 2475 METROPOLITAN BANKING ~AO 26
22 22 790q PERSONNEL GIO 82
23 23 1331 NURSE 020 lq 13 FLOOR #It#TRANSFER lq
2q 24 5161 CREDIT HIO 53
25 25 2534 S P E C I A L LOANS H20 26
26 26 1859~ PROPERTIES,SOUTH IlO 192
27 27 2739 SECOHITI I20 28 10 FLOOR ill,TRANSFER 28
28 28 1797 GOVERNMENT AFFAIRS JI0 18
29 29 1302 TRAVEL & RESERVATIONS J20 13
30 30 1980 COnTrIBUTIOnS J30 20 10 FLOOR VII,TRANSFER 20
31 31 5927 BOND 6 M O N E Y M A R K E T RlO 56
32 32 7359 OPERATIONS 6 TELECOEM. Lt0 76
33 33 762 NINE T R A N S F E R i20 8
39 34 4310 C.I.R.C.O. M10 91
35 35 2720 CORPORATE FINANCE NIO 28
36 36 2986 CCRE~L. S E R ¥ I C E S PIO 31
37 37 5054 C H A I T E L P A P E R - L E A S E FIN. F20 52
38 38 1939 LEASE P30 15
39 ]9 1750 LEGAL P~O 18
90 ~0 3237 ROLlS.RECeIVAbLE P50 33
41 ~1 1727 E X A M ! N A T I O n S . SOUTH P63 18
42 42 4392 COMNL.SE~VICES-ADRIN. P70 45
~3 43 &~596 FACTORS PRO 25q
~ qq 923 DISTRIBUTICZ Q19 13
95 q5 3911 NORTGAGE $10 qO
96 q6 863 E X E C U T I V E O I l I N G - O F F I C E S T10 9 1~ FLOOR |11.TRANSFEE 9
97 47 2587 PERSONAL F I N A N C E C E N T E R 510 27 FLOOR ,2 PLAZA 27
~8 96 8506 MAIN ~ L E V E L CNE~ 52A 88 1 FLO08 .2"PLAZA 88
49 ~9 19137 RAIN + L E V E L T R G ~ 52B 198 ~ FLOOR ,3 : E E Z Z A N I N E 198
50 50 22956 EXECU IVE FIGO 53 530 237 19 FLOOR R53 237
51 51 17298 E X ~ C U T I V E ~ FLOC~ 5~ 5~0 179 20 FLOOR #54 179
52 52 2122 EXECUTIVE DINING 550 22 29 FLOOR N5R 22
53 53 2668 BROKE bAG L U N C 8 ROOM 580 28 10 FLOOR |11,TRARSFRR 28
54 5~ 1390 CENTRAL REPRODUCTION 590 11 13 FLOOR All,TRANSFER 11
---~:T~---~,~~u~:j~zz, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 6 9 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .861
..
b ~ OF TOTAL SPACE A V A I L A B L E : 97Y ~ OF S P A C E P R E A S S I G N E D : 22~

Figure 2. 5taching problem (multi-storey banh) (a) zone definition (b) octivity definition

volume 13 number 5 september 1981 279


to desired travel time). Straight-line distances are suitable
in contexts such as open office floors, where it is usually
possible to walk fairly directly from one location to another F- I t
(Figure 4e). City-block distances are suitable in contexts
such as partitioned office floors, with corridors, where
pedestrian movement takes place mostly in orthogonal
directions (Figure 4b).
In the stacking plan example, inter-floor travel times
were entered directly as shown in Figure 5a. Values indicate
travel time in seconds, which was estimated on the basis of
actual walking and elevator times.* Straight-line distances
between zone centroids were used in the block plan
Figure 4. Types of distance measure
example (Figure 5b).

*A planned addition to the SpacePlanningSystem is a routine I nteraction data


which would produce travel time estimatesgiven elevator banking Interaction values can be given on a subjective scale repre-
information, senting judgements concerning grouping requirements, or
ZOgE DSFIIIZSlOg may represent an actual measure of flow of goods or
tgllt t o~
1D! (aeegL IS ) DH~RI PTIga nODULES employees. The program associates the interaction value
g leg specified for a pair of activities with each pair of individual
.................. .so~_~.........................................112
SOOTH-VEST
_~_~_ activity modules. For example, for any two activities, one
composed o f p modules and the other o f q modules, there
SOD| | | s g
are actually p x q pairs of activity modules which are
assigned the activity-activity interaction value. In order to
..,~, i~ ............ i;~ ...................
neutralize the effect of activity size on the objective func-
211-
tion, each interaction value is normalized by dividing by
2o! the sum of the number of modules associated with the two
lg/
1e 1-
activities.
17! Since area constraints may force a single activity to be
split between one or more zones, a large 'split penalty'
'~'i
15 , ......................................... ............... _-:i .................
interaction value is associated with pairs of modules belong-
,sj ing to the same activity. This penalty value then acts to
insure assignment of the activity parts to adjacent zones.t
':i 1°'
'g!
]i°3 Figures 6 and 7 respectively show the interaction
s I- .! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . matrices used in the stacking and block plan examples.
J In both cases the data represent subjective judgements
made by the client. In the block planning problem, the
.!
¢
2! tFrom experimentation,the ,split penalty' is set at three times the
largest value in the interaction matrix. This valueseemssufficient
oI ......................................... , ................................... ,
to ensure adjacency requirementsbut not so large that all the
a ~-V" !--*.--/,--V~V-V-,VW, V,F~,;I-~-~V~,;I,-;~:~6~.l-i]-~J-~;~-~ other interaction valuesbecome insignificant.

ACTIVIT! DE~IJ/TIO N
£EEA J OF COO PROP P R E A S S I G N E D TO ZONE: # OF
IDA (SO F R E T ) DESCRIPTIOH nODULES RATIOS ID| DESCRIPTION HODULES
...................................................................................................................................
I 1 6200 O R I E N T A L BAZAAR 62
2 2 1100 RUGS 11
3 3 2300 FLOOICOVERING 23
~ 1100 SILV|HH&RE 11
1390 SILV|R SHOP 14
~ 1500 GLASSHARR 15
~8 o . s s ~oP
8 8 BAD ACCESS. •
9 9 1500 rile CHIn 15
I0 10 2RO0 CISOAL C H I N A 24
II I1 2250 PLACBflATS 23
12 12 lq60 PICZBIBBONS 15
13 13 2650 LAHPS 27
14
15 is
14
j~
1170
~OeSLs
SUADES T2
16 16 . . sHoP ,~
17 1, 15o . , . ~oP 2
18 10 1~ 50 LIBELS 15
19 19 760 BATH BOOTIORE S
20 20 7~0 G I E ~ SHOP 8
21 21 17 0 PILLOH S H O P 17
22 22 lqO0 BL&|KRTS 14
~
24
23 400 sPgs60ssRoP
24 2500 SHE I T S 25
t~ 25
26
17~8 oEco, P.RTS
N&LL PAPER
1~
!i
.oDo s, ,. ,. .oss
.
31
30 30 DECOR. SHOP 12
--'i~T'-~; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4.7.8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . O
I OF T O T & ~ S P A C E AVAILABLE: 95R % OF S P A C E P R E A S S I G N E D : ~

ACTIVITY BEIASSZGNBENTS
# OF P R E A S S I G N E D T O ZONE: I OF
ID
..... .8
. . . . . . . . . . . . . . . .D.B.S.C.R.I.P.T.I.G.E. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .N.O.D.U.L.E.S. . . . . . . . . . . . . .ID|
. . . . .D.E.S.C.R.I.P.T.I.O.N. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . H. O.D.U L E S
1 O ~ I R N T A L BAZAA~ 62
..................................................................................................................................
2 NORTH 62
TOTAL: 62
b N OF SPACE P R E A S S I G N E D : 12~

Figure 3. Block plan problem (department store) (a) zone definition (b) activity definition

280 computer-aided design


TRAVEL TI~E EATRIX

ID# ZONE CARD TRAVEL TIRES


. . . . . . . . . . . . . . . r-. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.
FFLOOR
L O O R I~,8E~ZZANINE
|2#ELAZA 51. O.
3 FLOOR 59. 53. O.
4 F L O O R #E 67. 61. 55. O.
5 F L O O R 16 75. 69. 63. 57. O.
6 FLOOR 17 78. 76. 70. 64. 58. 3.
F L O O R |E 88. 82. 76. 70. 64. 57, O.
FLOOR 19 94 80 02 76. 70. 55. ~.
9 FLOOR I1G lOO 9.. 88. 82. 76. ~: 61. 53.
9 F L O O R 110 2 O.
10 FLOOR E l l , T R A N S F E R 136. 10J. 9~. 88. 82. 75. 67. 57.
10 F L O O R | 1 1 , T R A ~ S F E R 2 51. O.
11 F L O O R t 1 2 192. 186. 180. 17qo 168. 161. 153. 145.
11 fLOOR 112 l 137. 51. 3.
12 FLOOR t 1 4 200. 194. 188. 182. 176. 169. 161. 153.
12 F L O O R t 1 4 2 145. 59. 53. 0.
13 FLOOR t 1 5 208. 201. 196. 190. 184. 177. t69. 161.
13 F L O O R 115 2 153. 67. 61. 55. O.
14 FLOOR | ! 6 215. 209. 203. 198. 192. 185. 177. t69.
14 F L O O R #16 2 161. 75. 69. 63. 57. O.
15 FLOOR t 1 7 223. 217. 211. 2~5. 200. 193. 185. 177.
15 FLOOR #17 2 1 6 9 . 8 3 . 77. 7 1 . 6 5 . 5 9 . O.
16 FLOOR 119 235. 229. 223. 217. 211. 204. 196. 189.
16 FLOOR 119 2 181. 95. 89. 83. 77. 71. 65. O.
17 FLOOR # E l 288. 296. 304. 312. 320. 326. 332. 336.
17 FLOOR /51 2 344. 334. 342. 373. 365. 365. 368. 330.
17 FLOOR I ~ 1 0.
18 FLOOR 1~2 1 290. 298. 306. 314. 322. 329. 335. 339.
18 FLOOR 152 347. 336. 344. 376. 364. 367. 371. 382.
18 FLOOR I ~ 2 3 51. O.
19 FLOOR #~3 293. 300. 308. 317. 324. 331. 338. 341.
19 F L O O R 153 2 3 4 9 . 3 3 9 . 3 4 6 . 3 7 8 . 3 6 6 . 3 6 9 . 3 73. 384.
19 FLOOR t E 3 59. 53. O.
20 FLOOR | 5 4 1 295. 303. 311. 320. 327. 33~. 340. 345.
20 FLOOR | r q 351. 341. 349. 381. 369. 371. 375, 387.
20 FLOOR 1~q 67. 61. 55. O.

TBJ, VEL T'Y8 R BRTRIX: CALCULATED 15 STRAIGHT L I R E DISTARCE


ID| ZOllm CARD 8 TRAVEL TIBER
I RORTB-NEST 1 0
2 RORTB 1 15. O.
3.0ItTn'~|BTRRL 1 13. 8. 0.
4 SOOTB-' RST 1 8. 70: 15. O.
5 SOUTH 1 15. 8. 13. O.

Figure 5. Travel time matrices for stacking and block plan problems (a) interfloor (b) interzone

client specified adjacency needs on an ordinal scale activity of 3000 ft 2 yields a cost component of 300.
(1 - most important, 2 - less important, 3 - least No fixed costs were used in the block plan problem
important), and these values were translated by the program as the major emphasis was on generating configurations
into points on a ratio scale (8 - most important, 4 - less which best met adjacency requirements.
important, 2 - least important). The program provides
facilities for translation of various kinds of scales used for Discussion o f data
input into a ratio scale as required internally. Actual space planning problems of the type under considera-
tion may involve data that are objective or subjective, and
Fixed costs reliable or unreliable.
The fixed costs can be entered either as actual, detailed A factory floor layout problem, in which interactive
cost values for each activity-zone combination or can be costs are derived from measurement of materials flow, and
entered via a special simplified code. fixed costs are good estimates of costs of moving items of
Actual cost values might represent rent or special equipment to various locations, is an example of a problem
facilities construction requirements. If an activity is already in which the data are objective and reliable. In this case the
located in a particular zone, the fixed costs of locating it value of the objective function may be interpreted as a
in another zone may include moving costs. measure of actual yearly operating cost (assuming that
The simplified mode of entry of fixed costs allows the fixed costs are amortized over some specified period).
three fixed cost options for each activity-location com- An office floor layout problem, in which the data
bination: (0) suitable for assignment, (1) not suitable represent somebody's intuitive expression of location and
for assignment or (2) assignment prohibited. Suitable for adjacency requirements, is an example of a problem in
assignment implies a zero fixed cost. If a zone is which the data are subjective. Essentially such data repre-
considered not suitable for assignment a S/unit cost must sent the informant's preference and indifference relations
be specified. While it is possible for an activity to be on sets of activity-location pairs, and sets of activity-
assigned to such a zone this cost is incurred if the assign- activity adjacency pairs. As such, they may exhibit various
ment is made. This is balanced against the interaction inconsistencies and biases. However, even quite unreliable
Costs in the objective function. When an assignment is data may provide a useful starting point for application of
prohibited, the program is constrained from placing an the Space Planning System when it is employed iteratively,
activity in the specified zone. as discussed later.
Figure 8 shows the fixed cost matrix used in the stack- Since disparate kinds of data are employed as input,
ing plan problem. In this case prohibitive fixed costs have and since any ordinal data must be translated internally
been used to place certain floors off limits to many into points on a ratio scale as real numbers, it is important
activities (eg floors 1,2, 10, 19, and 20). The given cost/ft 2 to maintain a critical awareness of the properties of the
is applied when a 'not suitable' category is specified. Note scales that are employed, and the effects of any rescaling
the selection of interactive and fixed cost values so that procedures that are applied. Otherwise formulations may
theyare reasonably balanced in magnitude. For example become unrealistic, and results may be misinterpreted.
an interaction between two activities of 10 multiplied by It is also very important to remain aware of the assump-
a travel time of 50 between two floors gives a cost tions that are built into the. objective function. For example,
component of 500. A fixed cost of $.10/ft 2 for an while very small savings in travel times between activities

volume 13 n u m b e r 5 s e p t e m b e r 1981 281


INTEBACTIO~ MATRIX

INN & C T I W I T 1 IN| ACTIVITY VALUE


...............................................................................................
i IRTEBRA~IGHAL,OPERATIONS HI0 I CORPORATE BARKING AI0 10.
ZRTEANA~IOR&L,N&BKETING B20 2 INTERRATIOIIAL,OPENATIONS El0 15.
TBBST,C||To~CCTG. C20 4 TBBST,RBRINISTBATIGR CI0 15.
6 EHPLOIE|EaPLOYE|TBgsTTRmST c30C30 ~ TBUST,ADBInISTRATIORTBDST,CENT.ACCTO. CIOC20 15.15.
CONS~RRTTION FINANCE D20 7 REAL ESTATE FIN. IRSTITo DID 15.
10 REAL ESTATE APRB.,INCOBE E20 7 REAL ESTATE BIN. IRSTIT° DR0 13.
10 REAL |SI|TE &PPR.,INCOHB E20 8 CONSTRUCTION FINANCE D20 19.
12 COaHEECIRL ~OBEGAGE El0 10 DEAL ESTATE APPR.,INCORR E20 I0.
~ BRANC~ AERINIST RATION F,0 tq BHANCH PLANNING F30 13.
ADflIBIS~BATION Eg0 14 BRAHCH PLANNING F]0 10.
20 A D ~ | I R I B A T I O N F90 15 BRANCH ADMIRIST RATION Fq3 13.
23 NORSE G2C 13 DISTRICT 6 F23 13°
2~ C B E D I T H10 23 NURSE G20 10.
25 S P E C I A I LCANS 1120 1 CORPORATE BANKING A10 1}.
27 S ~ U R I T Y 120 I~ BRANCH PLANNING F30 I~.
~ CONTBIRR~ION5 J30 29 TRAVEL ~ RESERVATIONS J2O 10.
BOND S BURET NARKET El0 29 TRAVEL E RESERVATIONS J23 15.
~ BOND & mONEt NARKET ~I0 30 CONTRIBOTIONS J30 t0.
OPBRA~I£RS & TELEC£RH. LI0 5 TRBST.CEMT.ACCTG. C20 10.
35 CORPORATE F.IRANCE N10 2 INTERNATIONAL,OPERATIONS B10 2J.
35 C O R P O R E A L FINANCE RIO 3 INTEBN&TIORAL.NARKETIRG B20 I0.
36 C O ~ B e I o ~ E R V I C E S El0 1 CORPORATE BANKING A10 13.
3b COHBULoE|RV~CES PI0 32 OPERATIONS 8 TRLBCOHM. LI0 1}.
~ COBfl'L.$RRVICES PI0 35 CORDOBATE TINANCE NIO 10.
CHATTEL PAPER-LEASE FIN. R20 I CORPORRTE BARKING A10 |J.
37 CHATTEL PAPER-LEASE FIN. P20 3B COBNwL. SERVICES P10 I0°
39 LEGAL ~q0 38 LEASE P33 10.
q0 &COTS.RECEiVABLE P5O I CORPONATE BANKING At) 13.
~0 ACCTS°RRCEIVABLE P50 38 LEASE P30 10.
~2 COMMeI.$ERVICES-AE~IN. P70 38 LEASE P3~ 13.
q3 E&CTOBS P80 38 LEASE PS0 15.
~3 FACTORS E80 39 LEGAL P~0 10.
43 FACTORS fH0 ~O ACCTS.RECEIVABLE P53 1J.
q3 FACTORS P80 ~2 COBR°L.SRNVICES-ACNIN. PT0 10.
q3 F&CTONS [80 4q DISTRIBUTION Q10 15.
45 NOBTGA~E 510 I CORPORATE BANKING A13 I).
~5 8OETGAGE $I0 2 INTERNATIGN&L,O~ENATIONS B10 10.
45 flORTGRGE 510 3 INTERNATIO~&L,a&RRETING B20 13.
~5 NORTGA~E ~10 q TROST,ADNINISTEATION CIO I}.
q5 BOR~GAGR 510 5 TBUST,CRNT,ACCTG. C20 10.
~5 MOR~G£&E $10 7 HEAL ESTATE FIN. INSTZT. D]3 13.
~5 MORTGAGE 51E 8 CONSTRUCTION FINANCE D20 13.
~5 MORTGRGE 510 16 L I & B I L I ~ CARD FS0 10.
q5 MOBTGA£E 510 27 SECORITY 123 13.
~5 MORTGAGE 510 38 LEASE PSO 13.
q5 flORTGRGE 510 ~0 ACCTS.RECEIVABLR P53 I'3.
~5 MORTGAGE 510 q3 FACTORS P83 I).
~b EXECUTIVE DINING-OfFICES TI0 38 LEASE P]O 10.
~6 EXBCRTIYE DINING-OFFICES TIO ql EXAMINATIONS, SOUTH P63 13.
q6 E X E C N T I E E DINING-OFFICES TI0 ~3 FACTORS P80 I).
~ 6 BXBCR~IYE DINING-OFFICES TI0 ~5 MORTGAGE SIO 10.
q7 PERSONAE FINANCE C£~TEE 513 8 CONSTRUCTION FINANCE D23 15.
~7 PERSONAL FINANCE CENTER 510 7 REAL ESTATE FIN. INSTIT. DR0 15.
q8 B A I N | L E V E L ONE} 52A 33 CORTRIBUTIORS J33 13.
50 BXBCBTIYR# FLOOR 53 530 1 CORPORATE RANKING At0 13.
50 BXECB~IYE, FLOOR 53 530 CI0 10.
50 EKECR~XVR, FLOOR ~ 533 I~ TRUST#AB~INISTRATION
DISTRICT6 F23 I).
EIECRTIVE, FLOOR 53C 17 ESCNOR F60 I).
5u0 EKZCgII~E, FLOOR 53 530 25 SPECIAL LOANS H20 ~0.
EXECB~I~E, FLOOR 53 530 ~6 EXECUTIVE DINING-CFFICES TI0 15.
38 LEASE P30 ~6 EXECUTIVE DINING-OFfICES TI0 10.
38 LEASE P30 ~q DISTrIBUtION Q1O 15.
INTERNATIONAL,OPERATIONS DR0 ~ INTERN&TICNAL.NAPNE~ING B23 23.
TRDST,AEMINISTRATICN CI0 TRUST,CERToACCTG. C20 20.
TKBS~,CENT. ACCTG. C23 6 EflPLO~EE ~RUST C30 10.
REAL ESTATE FIR. INSTIT. DR0 8 CONSTRUCTION FINANCE D20 23.
9 REAL ESTATE APPB. eRESID. El3 10 REAL ESTATE APpN.,INCOHE E20 20.
10 HEAL ESTATE APpB.#IHCONE E20 11 AUDITING E3O 23.
12 COMNEEC]AL MORTGAGE Ft0 13 DISTRICT 6 F20 20.
t3 DISTRICT 6 R2O I~ BRANCH PLRRNIRG F30 20.
l q BRARCN ELANNING FSC 15 BRANCH ADMINIST ~ATION EgO 23.
15 BNANCR |UBIRIST P ~ I O ~ F~0 16 LIABILITE CARD FS0 20.
16 LIABILITY CARD F50 17 ESCROW F69 2}.
17 ESCROW FbO 18 LOAN-ADMIRISTRATICN F70 23.
|N LOAN-ADEINISTRATICN F70 19 INDIRECT LOAN-ADNIN. F8O 20.
19 I N D I N I C T LOAN-AUXIN. FBO 20 ADMInIStRATION F93 23.
20 A ~ I N E S T B A T I O R F~0 21 RETBOPOLITAR BANKING FA0 20.
22 P B E S C E E E L GI0 23 NURSE G20 20.
2 . CREDIT H10 25 SPECIAL LOANS H23 2J.
26 PROPERTIES, SOUTH II0 27 SECURITY I20 23.
2~ GOVEBSEENT AFFAIRS JlO 29 TRAVEL ~ RESERVATIONS J20 20.
29 TRAVEL R RESERVATIONS J20 30 CONTRIBUTIONS J33 23.
32 OPERATIONS £ TELECCHS. LI0 33 WIRE TRANSFER L20 20.
3b COMM~L. RERVICES P10 37 CHATTEL PApER-LEASE FIN. P20 20.
37 CHATTEL PAPER-LEASE FIN. P2C 38 LEASE P3D 20.
38 LEASE PSO 39 LEGAL P~O 20.
39 LEGAL Pq0 ~0 ACCTS.RECEIVABLE P50 29.
qO ROOTS.RECEIVABLE PSC ~1 EXAMINATIONS, SOOTH P60 23.
~I EXABIE&~2ONS. SOUTH P60 q2 CO~M~L. SENVICES-A~BIN. PTO 20.
~2 COM~eL.EERVICES-ACRIN. P70 ~3 FACTORS PH0 20.
q3 FACTORS P80 ~ DISTRIBUTION QiO 20.
36 COMMeL.SEBVICES P10 ~3 FACTORS PH0 20.

Figure 6. Activity interactions for stacking plan problems


on an office floor may be of no practical significance, problems. (See Garey and Johnson 3 , Karp 4 , Lewis and
summation of many such savings in computation of the Papadimitriou s and Sahni et al 6 .) It can be shown
objective function may have a considerable effect on its that quadratic assignment problems belong to a class of
value. (One possible response to this particular difficulty mathematical problems known as NP-complete. It is
is judicious use of thresholds and rounding off.) generally accepted that efficient solution of NP-complete
problems is impossible in principle. However, it is possible
to develop approximate solution strategies that produce
SOLUTION STRATEGY high quality solutions to realistically-sized problems at
In principle it is possible to solve space planning problems acceptable cost. Such solutions usually cannot be mathe-
as considered in this paper by exhaustive enumeration of matically proven to be optimal with respect to the specified
all possible ways of assigning activities to locations, and objective function, but given the nature of practical space
selection of the plan which yields the best value for the planning problems, this is not necessarily a severe dis-
objective function. In practice, for problems of realistic advantage.
size, this turns out to be an infeasible approach, since the Approximate solution strategies that have been developed
numbers of activity/location combinations involved are are of two basic types, known as iterative improvement
so vast. This remains true even when the largest and fastest strategies and constructive initial placement strategies.
computers are used. An iterative improvement strategy begins with some initial
Recent theoretical studies of computational complexity arrangement and attempts to improve it by exchanging
have thrown light on the difficulties that are involved in activities between locations. A constructive initial place-
efficiently finding solutions to large quadratic assignment ment locates activities one-by-one, to build up a solution
problems, and similar combinatorial optimization from scratch in a step-by-step fashion.

282 computer-aided design


INTEHACTIC| MATRIX

.VALUE
. . . . . . . . . .SCALR:
.............. CLD NEg
1 B
2 4
3 2
IN| ACTIVITI IO# ACTIVITY VALUE
UOGS I PLOORCOVEBING 8
RUGS 2 BEDSPREADS 6.
6 SZLVIRSSBR S SILVER SHOP 8.
6 SILV|RuARR 6 GLASSNAEE 4°
5 SILVES -cROP 6 GLASSNANE ~.
6 GLASSgAIR 7 GLASS SHOP 8.
6 GLASSS|IR 8 BAR ACCESS. ~.
7 GLASS S I C P 8 BAR ACCESS. 6.
7 GLASS SHOP 5 SILVER SHOP 2.
9 FIRm C S I | A 10 CASUAL CHZNA R.
9 PIER CHINA 7 GLASS SHOP ~.
19 BATH BOUTIQUE 11 PLACEHATS 2.
11 PLACRBA~S 9 PINE CHINA 6.
12 PIC/BIEIORS 13 LARPS 6.
13 LANPS l q SHADES 6.
i~ TOHHLS 16 BATH SHOP 8.
TOSELS 19 BATH BOUTIQUE 6.
L I B E l SBCP 18 LINENS 8.
17 LINER SNCP 16 BATU SHO~ q.
18 LINERS 16 HATH SHOP 6.
19 BATH HCRTIOOE 16 BATH SHOP 4.
21 PILLON SHOP 22 BLANKETS 8.
21 PLLLON SHOP 27 DRAPES ~.
21 P I L L O i SHOP 28 CONTAINS 2.
• 221 RLANEE'E-e 23 SPPHADS SHOP 8.
ELlllKR'I l 26 SHEETS 6.
2~ EL&|N|TS 17 LINEN SHOP 2.
SPREADS SHOP 2 6 SHEETS q.
I~ SHEETS 27 DRAPES 2.
DECOR PAINTS 26 HALL PAPER 8.
25 DECOR P&ZHTS 27 DRAPES ~.
25 DECOR |lISTS 28 CORTAINS 2.
t~ gALL P&PNR 27 DEAPES q.
DRAPES 28 CURTAINS 8.
26 BALL SIEUR 28 CURTAINS 6.
29 REDSPBEADS 28 CURTAISS 8.
I~ DBCOB. EBOP 29 BEDSPREADS 8.
DP.CONo EBOP 26 HALL PAPER ~.
30 DECOR. SHOP 25 DECOR PAINTS 2.

Figure 7. Activity interactions for the block plan problem


PIXED COST NATRIX 0 = SUITAULE
ZONE RESTRICTIONS: I = NOT SUITABLE
2 = PROSIBITHD
COST/ ;;;;;;;;;~;;11]];;ii ...................
EDA A C T I V ] ~ C&R~ SQ POOT 123qS678901236567890
1 C O R P O | A ~ R OAHKING klO 0.19 22000000020303000022
LSTES|ATZONALeOPERATIOHS HI 0.10 2220300002 0000002222
IIPEHRRA]ZONACeSARKETTNG H2~ 0.10 22000000020000000022
IROST,ATBZNISTRATICH C10 0.20 22000000020000000022
TBUST,CIRT.ACCTG. C20 0.20 22011111121111111122
5 REPLO|SI TRI~T C30 0.10 22000000020000000022
7 REAL SSIATE P I N . I N S T I T .
DI0 0.10 22000000020000000022
8 COISTRRCTIOI FINANCE U20 0.10 22000000020000000022
] 9~ REAL ESIATH A P P R . , B L ~ I D .
EIO 0.13 22000000020000000022
REAL ESTATE AP~R.eLHCOHE E20 0.10 22000000020003000022
AODITING ESO 0.10 22000000020000000022
12 COBBENCIAL HORTGAGE Yl0 0.10 22000000020000000022
13 DISTSAC~ 6 720 0.10 22000000020000000022
16 DRAICfl | L A S I I Z I G 730 0.10 22000000020000000022
15 BRABCE A C E [ l I S T HATLONYqO ~.10 22000000020330000022
16 L I A R I L I ~ ! CARD FS0 0.10 22000000020000000022
17 RSCBOb F60 0.10 22000000020000900022
18 LOEB- | £ H I U I STBA TION 770 0.13 22000000020030000022
19 IRDINRC~ LOAN-ADHIN. F6C 0.10 22000000020000000022
20 ADILI|LS~RATION FRO 0.10 22000000020000000022
2 1 E ~ T R O P O L I T A H HAUKIHG
FAC 0.10 22000000020000000022
t i PRRSC|R,L GIC 0.10 22000000010000000022
CRRDII HIO 0.10 22000000020000000022
25 SPECIAL LOANS H20 0.10 22000000020000000022
26 PROPRNIIESeSOUTH I10 0.10 22000000020000000022
28 GOVRRIBEST A P P , I R S JlO 0.13 22000000020000000022
2, T..L , RESHEVA~OHS
.O 0.10 22000000010000000022
~1 S O . , , 0 . , U.RRT
E'O 0.13 22000000020000000022
32 O . . , . , S , TELR~.,.
LIO 0.10 22000000020000000022
~3 . R , ..S.R L20 0.10 22000000020000000022
3q C . I . B . C . O . H10 0.10 22000000020000000022
3s PINA~CR
CORPOSASS Eio 0.13 22000000020000000022
]~ CORS'L.SnVICES rlo 0.10 22000000020000002222
CHATTNL PAPER-LEASE FIN. P2C 0.10 22000000020000002222
38 LEASE P30 0.10 22000000020000000022
39 LEGAL P~O 0.10 22000000020000000022
60 ACCTS°B|CRIVAULE ~50 0.10 22000000020000000022
61 gIAB INA5 [ONS SOUTH P60 3.10 22000000020000000022
62 CORReL.SERVI~ES-&CUIN. PTC 0.10 22000000020000000022
63 FACTOSS P80 0.23 22000000320000000022
65 BORTG~E $10 0.13 22000000020000301122

Figure 8. Fixed cost matrix for the stacking plan problem

The Space Planning System employs a sophisticated activity to a specific location and the restrictions this
constructive initial placement strategy based on an imposes on possible choices for future assignments. The
algorithm developed by Graves and Whinston 2 to create potential future cost of an assignment due to these restric-
an initial arrangement, then applies a simple iterative tions is the crucial factor. Since it is not feasible to
improvement strategy to refine the details. This coupling enumerate all possible assignments to determine exact
of a constructive technique with an improvement procedure future costs, probability theory is used to predict the
has been shown to result in a powerful solution approach effects of a particular assignment, thus enabling the
both in terms of the quality of solutions and the cost of algorithm to choose that assignment which seems most
generating solutions 7. likely to lead to an optimal solution. Specifically, at the
kth step of the decision process, the expected value of the
objective function is calculated by considering each feasible
assignment. The activity-location pair yielding the
Constructive placement stage minimum value is selected for the kth assignment. (See
At each step of a constructive procedure an activity is Graves and Whinston 2 and Liggett ~ for the specification
assigned to a location. The unique feature of the strategy and derivation of the expected value functions.)
developed by Graves and Whinston is the method for This solution strategy was implemented in the Space
selecting this assignment. The effects of a particular assign- Planning System with modifications to handle a number
ment are two-fold: the immediate costs of assigning an of special features of space planning problems.

volume 13 number 5 september 1981 283


Responding to area requirements
4
Once an activity-zone pair yielding the lowest expected
value of the quadratic assignment objective function is
selected for assignment, there are three possible conditions
with respect to area requirements:
• There is enough available space in the zone to
accommodate the entire activity. Minimum boundary rectangle
• Only a portion of the activity will fit in the zone, but 7 x 4 modules

there exist other zones which could take the entire


activity.
• Only a portion of the activity will fit in the zone, and
no other zone is available which could accommodate
the entire activity.
Only under the third condition must an activity be split
between zones. In this case as many modules belonging to
the activity'as possible are assigned to the selected zones
Proportion ratiO = 4 / 7
and the remainder of the activity's modules are considered Coherence ratio = 19/28
for assignment at the next step. At this point, the high
penalty interaction values between modules of the same Figure 9. Definition of shape constraints
activity help to ensure placement of the remainder of the
modules on the closest possible zone(s).
When the second condition occurs one of two options
can be invoked. The activity-zone assignment can be Responding to shape constraints
prohibited and only assignments to zones which can accept Because of area limitations and the nature of the construc-
the entire activity are considered. This option is used for tive procedure, irregular activity shapes are frequently pro-
the stacking plan problem where it is desirable to have as duced by the placement procedure when generating block
few activities as possible split between zones. This considera- plans. This is usually regarded as undesirable, even though
tion overrides adjacency considerations. On the other hand, such shapes might result in more efficient square footage
in the block planning problem, zones create artificial utilization. To force the generation of more regular shapes,
barriers and there is generally no reason a single activity the Space Planning System provides a facility for the user
should not be assigned to adjacent zones, since it can be to specify shape constraints.
joined at the boundary. In this case a second option, The concept of a bounding rectangle drawn around a
identical to that used at the third condition is invoked. shape is employed for this purpose (see Figure 9). Two
The first condition is the most desirable situation, simple ratios are calculated:
in which the best assignment can be made with respect to • the proportion ratio of the bounding rectangle, defined
interactions with other activities, while meeting the area as the ratio of the shortest side to the longest side
requirements. Two input parameters allow for flexibility • the coherence ratio of the shape; defined as the ratio
in the area requirements, thus increasing the probability of the number of modules in the shape to the number
that this condition will occur. These parameters are called of modules in the bounding rectangle
activity-compression or zone-expansion percentages. If a
zone-expansion percentage is specified, the area of a zone Shape constraints are specified by stating minimum allow-
can be expanded by the specified percentage if needed to able values for the proportion and coherence ratios for
ensure the assignment of an entire activity (note, zone each activity. The program attempts to satisfy stated shape
expansion is not allowed when specific zone perimeters constraints, but location requirements are always allowed
have been given as in the block plan example). Alternatively, to dominate, so in some cases certain of the shape constraints
the area of an activity can be compressed by a specified will not be met.
percentage if needed to fit into available space on a zone.
If a fit cannot be made even after applying these parameters, Ordering assignments
the second and third conditions must then be considered. The decision process of a constructive procedure may be
either 2D or 3D. In a 2D procedure, the order in which the
activities are selected for assignment is fixed and at each
step only possible locations for a single activity are
Responding to assignment constraints evaluated. In a 3D decision process all unassigned
As noted earlier, the fixed cost matrix can be used to activity-location combinations are evaluated at each step.
introduce constraints into the assignment process by While 2D techniques are less expensive computationally,
specifying prohibitive costs for particular activity-zone 3D techniques generally yield better solutions.
combinations. These prohibited assignments are then The Graves-Whinston algorithm is a 3D procedure.
ignored during the construction of a solution. However, options in the Space Planning System allow a
Constraints may also be imposed by using the option of modified 3D approach for intelligently limiting the
preassigning activities to zones. The preassignments of assignments evaluated at each step.
activities, which are specified in the activity definition input At the zone assignment level, the activities to be placed
data, are made in the first steps of the decision process. If can be sorted before assignment either by level or inter-
an activity is larger than the zone to which it is preassigned, action or by area. At the individual location assignment
the remainder of the activities modules will be placed by level, activities can be sorted by coherence or proportion
the constructive algorithm. ratios. One of two placement options can then be invoked,

284 computer-aided design


each requiring the specification of a group size N. Under
the first option, the top N activities in the sorted list are
placed first before the remainder of the activities can be
considered for assignment. For example, sorting by area
and forcing the placement of the largest spaces first can
result in fewer activities split between zones.
Under the second option, a cascading group approach
is used. After members of the initial group have been
placed, the next N activities on the list are considered
for assignment and so on. If the group size specified is
one, this method reduces to a 2D decision process. This
option can be used in combination with sorting activities i

by shape ratios in order to place those with more rigid


shape requirements first, thus ensuring they will be
satisfied.

Improvement at the zone assignment level


After an initial plan is generated by the constructive
procedure, there is usually room for local improvement. Figure I O. Shape improvement
The improvement procedure used is based on the simple
pairwise exchange principle. For each activity, a move
to each of the other zones is examined. If a unilateral expected effect of an individual activity location decision
move would improve the value of the objective function, on overal] efficiency.
the routine investigates how this might be accomplished. The strategy also recognizes that constructive initial
If there is available space in the zone, the activity is placement strategies and iterative improvement strategies
simply moved. Otherwise possible pairwise exchanges should not be regarded as rival approaches to the solution
with activities in the target zone are attempted. If no of shape planning problems, but rather as necessary
exchange which meets the area constraints and improves complements to each other. Since the optimization
the overall value of the objective function can be made, problem is NP-complete, the task that can realistically be
the activity is left in its original position. This process is addressed is one of finding a very good local optimum,
continued until a local optimum is reached (ie no further rather than the global optimum. The implicit enumeration
exchanges can be made which will improve the value of procedure provides a reliable way to get near to ah excellent
the objective). local optimum, while the improvement procedure moves
Once a local optimum is reached, a form of sensitivity the solution right to that local optimum. The result pro-
analysis can be carried out. Possible shifts of activities duced by using a constructive initial placement procedure
which would increase the value of the objective function only usually leaves room for improvement by small
within a specified margin are reported. This provides the readjustments, while the result produced by iterative
planner with optional configurations within approximately improvement of an arbitrary starting solution usually
the same cost region. yields one of the poorer local optima.
The multi-stage application of the Space Planning
System combines the advantages of top-down and
Improvement at the individual module level bottom-up design strategies. Multi-stage application is a
Activity shapes generated at the block plan level may be classical top-down strategy in which, at each successive
irregular, even with the application of proportion and stage, a complete solution at a higher level of resolution
coherence constraints. A final improvement process aimed of the details is produced. Cohversely, within each stage,
at modifying shapes exchanges individual activity modules constructive placement of activities amounts to a
if the exchange results in an increase in the coherence of bottom-up approach to a solution by solving a sequence
one activity without decreasing that of the other. Figure 10 of small sub-problems.
shows an example of this simple process.
OUTPUT
Discussion of solution strategy Output from the Space Planning System can either be
A distinctive feature of space planning problems is that displayed at a low-cost, low-resolution colour raster
both an overall requirement (efficiency, as expressed by terminal (which is quite adequate for the simple graphics
a low value for the quadratic objective function), and a involved), or else produced on a line printer or a plotter.
variety of detail requirements (eg for adjacency between
given pairs of activities, or shape of the space allocated to Stacking plan output
a given activity) must be satisfied. In a step-by-step activity Figure 1 la illustrates part of a stacking plan generated in
placement procedure it is relatively easy to assure that response to the example data. It consists of a list of activities
detail requirements are satisfied at each step (at least for assigned to each floor, and a comparison of the total area of
the activities that are placed fairly early in the process) but activities assigned to each floor against the total available
it is often difficult to assessthe overall implications of an area. Expansions and compressions are flagged. Figure 1 lb
activity location decision. The Space Planning System's shows an analysis of fixed and interactive cost components
solution strategy deals with both aspects of the problem for the solution. An overall cost is given, followed by an
by combining simple decision rules that respond to detail activity-by-activity cost breakdown. Figure 1 lc shows the
requirements with use of probability theory to assess the sensitivity table that is produced.

volume 13 number 5 september 1981 285


I~lI
l f f t ~ l l I I I I • I I B I I I I I I m l I I I I I I I I U I I U ~ It• 8

JLOOI 85~ (A 20)


|zecuzl~s, rLOoi SR 16o~ ~ .~ns
...... =.... ........
~ ................................................
AREA A S S I G I I ~ D 1 8 0 ~ Sg R T R S
ARiA AVAILABLE 1 9 8 ~ SU N T R S
.................................
DIFFERENCE 180 EQ R T R S
IIIIllIm~II•immImmnmmsmm•tmummsmmm--mmmmsm====::=====~=:mm==:====: :===:~

/I,ool 06
0...0.6...... .... ~--.. ...... . ........................
11 51 -- . . . . . . . . . . . . . .

~~ ! ~ . ~
RURToJ~RZ|ISTEA~IOR
..............
C10 (A q)
[ ~ . . . . . . . . . . . . . J L / ~ I . . . . . . . . . 1!1]_~_~[~.
b9J 5~ ~TNS

A R E A ASS~fiNED 1866 SQ MTNS


.ABEA
. . . . .A.V.A.!.L.A.B.L.Z. . . . . . . . . . .1.8.6.2. . .5Q
. ~TR~ ***e A~ZVZT! AREA BET BE C O M P R E S S E D R! 0.22 O|
DIFFERENCE -Q SO RTRS Zo|E AREA RUST BB EZPAEDE8 BT 0.2S
I t I l l ~ l l l l l l l l l m t l l l l t t l l I t l l l l l l l l l m l l l l l l l l l l l l l l l l l = = = = l = = l l l l

ILOOE R5 (l .}

TEUSImC|RT.ACC~G. C 0 156 ~TRS


EB~LC~EE TRUST C~O
O~EIAS|~S ~ T|L|CQI~ L! I 683
girl ~'AISF" L28 (i ~3) 70 ~ HT $
............................................................ ~Z~_
AREA A S S I G N E D 1865 SQ BTE5
AREA A V A I L A B L E 1662 £G STRS
.............................. =*** ic~Vto~ AREA Rust ss¢
z ¢oaet~SEO aT o. s
• DIFFERENCE -3 SO H T R S a|E& B U S T EZFA|DED B: O.~S OR
i I U I i I U ~ l l I I I I S U I • I m J t t i J t I l = •

e==o= ,. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . J!__]L . . . . . . . . . . . . . . . . . . . . .
~!~S~_~E..A¢5~h . . . . . . czo .......... !L_~L ........ !~2.~_~!~.
. . ASSI.ED l s o ~ ~0 E,RS
A. AVAIlABle ,E11 s o ,~Rs
DIFFERENCE 2 SG HTRS
~ t l l ~ t l l m l m • l l m t i f l l = l t l m l f l t t t i t t =: =: : :•= I t l i t = = l= = = == = ==Z = t

iq~OI |3mBIZ=ARIRB (I 21
DXSXn..XOR o,o ~, ..~ o s s o ErRs
u ~ _ . E 2 , p . . j ! o ! ......... s_=~. . . . . . . . . . . J!_~2l . . . . . . .
!ZZZ_~LgEL
AR. A~s~.9, ,862 ~o R~NS
•. .NEA ..LAB~E 1~ss so . T . s
...............................
DIFFERENCE 93 SO RTRS
I I N I N in I 8 N U N I N I I I IlJI JlI N I I I I K N I I I I It ~ t : : : = ~ 1 8 Imt t ~ i i : z l : : = t : : i t

~LOOI 12,}liJ& (I 1)
~l|Cl|| ||l|lCI ¢IITII 510 (l q71 2NO SO flTRS
BAIl {&IV n~ ORE) 52A 18 U8) 790 SO E T R S
.. Assi.E~ IO~O ~ RTEs
...Az~.~E ,I. .~gs
DXFFE EECE 103 S~ 8 T R S
I m B t l f f t l t B I t t l l t ¢ l l f l t l t i l l u jijj i • s:z~m~nt umsm~u• i = :~tz~=tm

Ig/t ,,,A ASS~ORED


AB~A AVAILABLE
tttll |~ .SsT RsS
a DKfFBEERCK e312 SQ BTRS ( t lS)
ee*e*eee**0e~*e*ee*ee~e~e*e****eee**ee

* COST O~ PLAN: 33H5~b.2~ *

Block plan output


BHEAKDOHH CX COST BY kCTIVITY
IH# ACTIVJ~T FZXHH COST I~TERACTIV~ COST Figure 12 shows output from the first stage of the two-stage
--~-cG~;o;i;F;;;~i;;
INTEBHA~IGMAL OP[RAT2OHS
........ ~;G ................... G'G .......
HlO 0.0
;1;;;-;;
10309.99
..... example block plan problem, in which activi-~ies are assigned
IRTERS&| |ON eL ' flAE KETI K~ b,O 0.O 5829.99
0.0 90879.50
TROSTmA£BIRIS~UATI£H
~Rgs'! C|ST. ACCTG.
C10
C20 5773.73 63719° 93 to zones. As with the stacking plan output it consists of a list
E ~ L O ~ ! T~OST C30 0.0 lq919.98
i REAL ESTATE FIU, INSTITo
COHST~DCTZOU F 1HA lICE
glO
D20
0.3
0.0
3860.00
3860.00 of activities assigned to each zone, but in this case there is
RE&L~
AL ZS~&TEESIATE NPPR°,HHSZD. HtO 0.0 0.3
to9
H
R
AUDI~IHG
APPI~., IHCO~E E20
E30
0.0
O.0
3420.00
2280.30
no expansion or comi~ression of activities.
12 COBJERCI~L MOHTGAG~ FIO 0.0 3H60.00
13 015~BIC~ 6 F20 0.0 17159.98 Figure 13 shows output from the second stage, in which
14 BRARCR |LRMHZHG ~30 0,0 12619.98
15
16
BR&BCB i£flZHIS~ RATION
L I A B Z L ] ~ I CArD
FQ~
F50
0.3
0.0
6259.99
8579.99 a detailed layout is generated. Note how some activities
17 ESCRO b Y60 0.0 9919.99
18
19
LOAM- JL~BZUZSTRA2ZOH
ZRD 'Re LORN-AC~I~.
FIO
'8
0.0 0.0 (for example 24 and 15) have been split between zones,
20 A ~ t l |StRATION Y 9~ 0.3
0.0 0.0
1220.00
21
22
BETBO|C[IT&R BA~KII, G
PRESOHH|L
FAO
GlO
0.3
0.0
0*0
2280.00
but have maintained contiguity at zone boundaries.
~3 ,HR. ~,2o o.o ,o599.~9 No shape constraints were applied in the run that generated
CRRDI~ H 1C O. 0 6680.00
25 5PECIRL LORES H20 O.0 15839.99
26
27
P~PlRTI|S
SECURI~! '
SOOTH 110
I20
0.0
0.3
13~ 9.98
15H 9.98 Figure 13, and ragged shapes for some activities are evident.
28 ~ V R R R B | R T AFFA2H5 dl0 0.0 20HO.OO
29
30
~R&VRL 8 RESERV~TIC~S
COMTE3HITIOHS
J20
JJO
125.911
0.0
q890°00
~020.00
Figure 14 shows the result of re-running the problem with
31 BOlD & ReBEl MARKET H13 O.0 q750o0C
32
33
OPER&~I~RS 8 TELECC~ff.
MIRE TUIRSFBR
LIO
L20
0.0
~)00
~580.00
0.0
shape constraints added. The effect on overall efficiency is
l~l C.|~N co
CO CiA~E YXSAHCE
,10
~13
0.0
0.0
o.o
3800.00 insignificant for practical purposes, but activity shapes are
!~ COEn'L.;eRV C~S P10 0.0 1~019.98
CRETTEL EAP~R-LEASE FIN. F20 0.0 2920.00
31
38 LEAS| PS0 0.3 15379.99 now more coherent.
39 LEGAL Pq0 0.0 2~OoOO
II0 ACCTS. J |CE2 YABLH ['50 0.0 8099.99
ql BXAEZRA$IORS SOUTH P60 0.0 31~0.00
ill COIINmL.||RVI~ES-AD~IN. P70 0.0 69 9.99
~."~uzsTazo,,3o,
,~cToMs o~ o.o
O.0 s,2~,.,3
33789.98 Discussion of output
~5 flORTG&G| ~10 0.0 ~0559.9H
~ R~ECOTXI| DIRLNG-OFFZC~-S TI0 0o0 16929.98
I%/ PERSOHA~ FZ,BARCE C~bTER 510 3.0 5280.00 The output produced by the Space Planning System is
U8 BAZ' ILIIVIel OR E l 52& 0.0 2120.00
.i, ..~
q9 ES~O IH~ FLOO 55
Tun 52H
530
~:~ 0.0 intended not just to display a single 'best' solution, but to
~I , R c u , z , C ,LOON ~, sun 0.0 s3~%~:~6
! ! NE ~ ,cuv-,,=.2..
, = , ~'CN =no. ~ssg 0.o°"° o.0°'0 provide a useful characterization of a c/ass o f very efficient
CEHTRML EEPROOUCTIC H 590 0.0 0.0
solutions. The particular solution that is displayed should
be interpreted in relation to the sensitivity data and cost
breakdowns that are provided. Usually these will indicate
~OSSIBLE A(SIYZTT EXCHENGgS WITfllR COST S Z S S I T I ¥ I T Y MAPGIN: O.
ACTIVZTY-2C|E WITH ACTIVITY-ZC~[ NEW COST how to construct numerous acceptable variants.
- - - - ~ . . . . . . ~. . . . . . . . . . ~ ...... ~ ..... ~]~;;~3;-
3~ 1~ 18 235110°06
3~ 1] 6 235110.J6

c IMPLEMENTATION
Figure 1 I. Output for the stacking plan problem (a) final The Space Planning System has been implemented on a
stacking plan (b) associated costs variety o f computers which range in size from an IBM 3033

286 computer-aided design


DEEIRTFlEBT ~ O R E then the number of stages can be increased. This reduces
the amount of core storage needed at each stage.
Z012
===========~ ================================= .=__=
. . . . . . .. . .. . .. . . . =. . .. . .. ... ... .. . .. . . .

,oE~.-,,ST C' 18 ........ INTEGRATION INTO THE DESIGN PROCESS


;~;~ ......................................
FLoo,m.,.zsG 1,i--i31.......... TGG-S2]oo
|~ FEETEE'~
+.. , . 1 70 ,.T
.EIT 0 2, +00 + "~ Experience with use of the Space Planning System in
..,t.o, I,.} +,o°~,.T
DECOI, SlOP
................................... ~ i ; - i ~| l i ~301 ........ ~G-~-+~-
1200 SO F B ~ practice has shown that it is most effective when employed
.&EEl
. . . . . . . .&T&~EBLE
. . . . . . . . . . . . . . . . . . . . . . . 9800 $Q FEET iterate°ely as follows:
D~FFERENCE q ] 0 SQ FEET
======================================================================
• Data initially provided by a client is employed for the
8OATS 1~ 2)
•=&°8~i~$;"ii~i~.=o~ ..........................
[.;--~71.......... ;-~;-T-;~?-%0
s~ FEn first run. Usually this data is incomplete, inaccurate,
B&I M;C~S~. 8 100 S FEET
TOgBLS
i
| 15 3 0 5 FEET and inconsistent in various ways, so the result that is
l ' t 8 ~ 8 880F
LZI|BS
# 17
I 18
150 S FEET
1~50 $ FEET produced is unsatisfactory. However, the relation
11201 760S~,ffET
~_~!~£ ............................................................
...8zogE° ~o61oso Fz~r between the data and the result is clear.
.E~ ~FS~.~E ~o~oo so FeET
DXFFEBEICE 190 SQ PEEr
• This experience prompts the client to make more
=====================================================================
requirements explicit, and to remove errors and incon-
I O E T.H
. .- C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .{ I. . . .3). . . . . . . . . . . . . . .
. .| I.I.I .I L -- .
SILERIiII' l q 110 S FEET
sistencies from the data, particularly by making more
S £ L F I I SBO, [l 5+ 139~ S~ FEET
GL&~S8 E
,:/,,rio,, 6 1500 $ ET
IIl 1121 ,,+0 88 lh, preassignments and by manipulating the fixed cost
TOllLS I O0 S FE~T
matrix.
&B|& k&VEZLAFlLE
SSIGIE° 7q 0 I ~ FFEET
88~0 EET
• This cycle is repeated a number of times, with each
iB8&

===========================
. . . . +;;;i~ii+i ........ +;+-;+-;+~
=========================================== successive problem more tightly constrained.
SOOTB-
. . . .8. |S~
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .( i ~}
PILLOE SHO~ (0 21} 1700 50 FEET
Gradually the problem is transformed from one in which
interactive costs clearly dominate the assignment process
|II I|' ,,o, I ~ l
0.~o8 ,...
.. ,.. . ~m
o= ,3oo ,o ,-.
"o' "" to one in which preassignments and fixed costs dominate,
o,,i. l, ++I +,0o I~ ,.T
C
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (. i. 28}
O I T I l I 8 1150 S~ FB~T and the computer program is merely adjusting details.
.. ,SSmEEO 11,8o so ,EET
&RE2 . . . . . . . . . . . . . . . . . . . . . . . .11200 S~ FEET
. . . . . . . . .~V&IL&BLE This approach recognizes that an acceptable spatial
D~FFERBNCE 60 5Q FEET
======================================================================= arrangement results from a complex pattern of trade-offs,
SOOT8
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ( I 5) .... and provides a structured method for making those
F,l. CiII&
c&SgAL CBI|A
li1~1
11
2qOO1500i~FEBYPEET
22 trade-offs. The result is not just an arrangement, but a
PL&CEBIT5 13 26~00 FE[CT
FEET
L&NFS
....................................................................
~EES SSSZ~.EO e800 SO EEET
justified and understood arrangement. That is, the reasons
&El& AV&ILIBLE 96~0 SO F E E T
................................. for location decisions are made explicitly, and possible
DIPFERESCE 800 $0 F E E T
================================== ..... ==================================
directions for acceptable variation are identified.
ToTEL &Eel ASSIGlE° q7330 SO rS~T
TOTkL ~ . . . . . . . . . . . . . . . . . . . . . . . .49800 SO FEET
. .E. .E. .l. . .AVAILIBLE
The Space Planning System makes use of both general
2470 SQ
°IPFEREHCE FEET
results of probability theory, and empirical rules suggested
by practical experience, as a basis for location decision
Figure 12. Zone assignments'for the block plan problem making. It can be used very simply, with default options
controlling rule application, but it is most effective if one
(a large mainframe) to a PDP-11/23 (a small mini), in both takes the time to learn how to manipulate the various
batch and interactive versions. The program is written in options, so as to bring different combinations of the
Standard FORTRAN IV and is easily transferable. Compu- empirical rules into play as appropriate to particular types
tation time is not a limiting factor. Problems of the size of problems.
reported in this paper require only a few seconds of CPU
time on the IBM 3033 and only a few minutes on the
PDP-I 1/23. Core storage requirements constitute the major CONCLUSIONS
limitation, but a multi-stage approach can always, in practice, While it is relatively easy for a skilled human designer to
be adjusted to overcome this. That is, if too much core produce reasonably good spatial arrangements, it is not
storage is required for solution in a given number of stages, so easy to make explicit the trade-offs and assumptions
, ..... +. . . . . . . . +. . . . . . . . . . . . . . . . . . . .
i,+ i+o io,

I l__r-+..t 10!+: __FL


L!i-:~ ........ 1

I: ............ L .I 1" .........


.1
li;~__+
08 07 15
.......................................... I--I .....I ................. .......................................... l°.eh_t_°_'+__! ~ .......... ,. . . . .
o3 I l +° + 128 I'41 °+ l jos
, ....... ::i , ................................
I .--P= I I I I I" ,~;t .......... | l I°' I I ~14 ( ~19 [ ~04 I
•-3 .--i-- .--L °-
) ] l._.+;t ...... 1 L....j,,= . . . . . 1£ . . . . . . . . . J I
+ i::r !--*' *i . . . . . ........ J "-=
........ i ..... +
I I
..................
~'m
m ....................
m
m ..... ~m
~.J
I '+ pE--i . - ....... 1~;---I '; ........ :-;;; ......... I~ I" 122 I I" l" l ++ ," I
I ..... l .--i . . . . . . . I 1 I-' • ..... '"~"~ ...... + ' ............ ' t
I 12+ I ........... 12+t--t l ! ........... ~ I r IT' t ........... '+++--. I .J .............. I 1
l ..... 1 l .--.+ P 5 /
..... l I. r-I I r *'° I
+t . . . . . I +' I--.
! 1. . . . . . . .
'+'+J ..... I,
r ,._.

Figure 13. Final block plan without shape constraints


• --t
i ;~ --i.__L
t ..... l
i .--t ~ +-J l
r -~
F/gum 14. Final block plan with shape constraints
- t . . . . . .
m. . . . . . . .

volume 1 3 number 5 september 1981 287


that form the basesof such arrangements, to understand REFERENCES
how they might reasonably be varied, or to see how well 1 Koopmans, J C and Beckmann, M ] 'Assignment problems
they might compare with potential alternatives. The Space
and the location of economic activities' Econometrica
Planning System not only produces high quality arrange- Vol 25 (1957) pp 53-76
ments very rapidly and at low cost (typically $1.00 to
$5.00 per solution), it also makes the basis for solutions 2 Graves,G W and Whinston, A 'An algorithm for the
explicit, gives sensitivity information, and helps rapid quadratic assignment problem' Manage. 5cL Vol 17
exploration of possibilities under different assumptions. No 7 (March 1970) pp 453-471
As such it is a very powerful tool for use in space planning 3 Garey, M R and Johnson, D S 'Strong NP-completeness
and management strategies that attempt to achieve such results: motivation, examples, and implications' J. Assoc.
goals as: Comput. Mach. Vol 25 No 3 (July 1978) pp 499-508
• satisfaction of the space needs of all employees and 4 Karp, R M 'Reducibility among combinatorial problems'
working groups, in Complexity of computer computations Plenum Press,
• enhancement of the productivity of the organization New York, NY, USA (1972) pp 85-104
by maximizing easeof communication, achieving 5 Lewis, H R and Papadimitriou, C H 'The efficiency of
cohesion of groups that must work together and algorithms' ScL Am. (January 1978) pp 96-109
achieving contiguity of closely interrelated groups,
• minimization of fixed operating expenses, particularly 6 Sahni, Sarlaj and Gonzales 'P-complete approximation
rent, problems'J. Assoc. Comput. Mach. Vol 23 No 3
• minimization of wasted and unused space, (July 1976) pp 555-565
• projection of and allowance for growth, reduction, 7 Liggett, R S An exploration of approximate solution
and rearrangement in spatial arrangements over time, strategies for combinatorial optimization problems
• comparison and evaluation of alternative plans and PhD dissertation, University of California, Los Angeles,
management strategies. CA, USA (December 1978)

288 computer-aided design

You might also like