Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

HUMAN FACTORS, 1967, 9 (4), 295-300

‘L

Linear Programming Applied to


Optimization of Instrument Panel
and Workplace Layout
LOUIS E. FREUND and THOMAS L. SADOSKY, university of Michigan, Ann Arbor,
Michigan

Thispaper investigates the aPPfication of several linear programming algorithms to the problems of work-
place and instrument panel design. The various approaches described depend primarily upon the type of
problem constructed; that is, upon the structure of constraints and available information. Two types of
optimization functions are described, one based upon the distance between available positions and the
second based on minimum eye travel. One formulation incorporates the probability of transition between
any pair of available positions as a design parameter. The results indicate that solution of these problem3
is possible by several algorithms,primarib the I ransportation algorithm and the assignment algorithm. A
solution using Simplex was attempted for one type of design and the constraint structure proved to be
complex.

The layout of a workplace or instrument panel OPTIMIZATION BASED ON


should provide for optimal operator performance. MINIMUM “UTILITY COST”
The literature shows a tendency to formulate gen-
eral human factors principles for workplace ar- Work Place Layout
rangement and to offer them as recommendations
to the design engineer. For example, the proper Consider the problem of optimizing the ar-
arrangement of displays and controls is a basic rangement of a workplace, for example an air-
problem in workplace layout. Among the princi- plane cockpit. Fitts (1947), investigatingthe prob-
ples cited for the design engineer are: (1) place- lem of pilot errors in aircraft control manipula-
ment of the most h p o i +antdisplays and controls tion, obtained data relating to the desirability of
in the optimum worklig and viewing area and various cockpit areas for the placement of con-
(2) organhation on the basis of function, fre- trols. In effect, he subdivided the cockpit into
quency of use, O r sequeace of operation. Effective eight distinct areas as shown in Figure 1. His data
application of such Pri iples requires that human reflected the relative accuracy of the pilot in a
factors criteria be exprr: e d in a quantitative form :blind reach to each of the eight areas. Subsequent-
useful to the design engmeer. ly, Deininger (1958). collected data on the fre-
The purpose of this Psper is to demonstrate the quency with wGch the pilot made certain task
applicability of several hear programming al- responses in flying simulated cargo missions. Com-
gorithms to the general problem described above. , : bining these two reports, it is possibIe to construct
The problem will be formulated in two different a “utility cost” rating for placing each of the eight
ways: (1) optimization of layout bas$dw @- controls in each of the eight areas where “utility
mum “utility cost” and (2) optirniktion of ia$&t cost” is defined as the product of the frequency of
based on minimum eye travel. Exajnple problems w e and the area mean accuracy score. (See Table
will be presented to illustrate the benefits or dif- 1) For example, the cost of placing control M in
ficulties of each approach. The specific algorithms area 1 is 3.49 x 13 or 45.37, and the cost of
used for solution may be found in sele$al,!inear Placing control E in area 4 is 3.13 X 30 or 93.90.
programming texts, such as Dantzig (1963). The problem is now in a form which can be
295

Downloaded from hfs.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on April 27, 2016
296August, 1967 HUMAN FACTORS

TABLE 1
Utility Cost Tableau (Fitts, 1947, and Deininger, 1958)
Controlst
Mean
Accuracy AP I E T C AC M C
Score Area 115** 40 30 21 20 15 13 12
2.14 246.10* 85.60 64.20 57.18 42.80 32.10 27.82 25.68
2.45 282.90 98.40* 73.80 66.42 49.20 36.90 31.98 29.52
2.76 317.40 110.40 82.80' 74.52 55.20 41.40 35.88 33.12
3.03 348.45 121.20 90.90 81.81' 60.60 45.45 39.39 36.36
3.07 353.05 122.80 92.10 82.89 61.40* 46.05 39.91 36.84
3.13 359.95 125.20 93.90 84.51 62.60 46.95* 40.69 37.56
3.21 369.15 128.40 96.30 86.67 64.20 48.15 41.73* 38.52
3.49 401.35 139.60 104.70 94.23 69.80 52.35 45.37 41.88.
tA bbreviations *Optimal solutions
Mixture (M) **Frequency
Throttle 0
Elevator Tnm (E)
Auto Pilot Panel (AP) creasing frequencies and the areas are arranged
Cross Pointer Set (C)
Gyro-Mag.Compass set (G) in order of increasing mean error scores. Table 1
Intercom Panel (0 has been arranged in this manner. The optimal
Auto. Compass panel (AC) solution, found by making assignments on the
principle diagonal of the tableau, is shown in the
Table by starred (*) numbers. A proof of this
method is given by Huebner and Ryack (1961).
Relying upon hypothetical data, we next con-
sider possible extensions of this problem as fol-
lows. Suppose we have c controls to assign to n
areas, where n < c, where each control occupies
equal space, but where the areas are of differing
size. As an example, let n = 3 and suppose that
there are capacities restricting the number of in-
struments that can be placed in a given area as in-
dicated in Table 2. We desire to place a total of 4
controls in the 3 areas. Table 2 shows the fre-
quency of use associated with each control. The
cell entries in Table 2 represent the product of fre-
quency of use and mean accuracy score. The prob-
lem is now in a form which can be solved as a
transportation problem. The solution is shown in

Figure 1. Workplace areas (Fitts, 1947) TABLE 2


Initial Tableau for Transportation Algorithm
Areas 1 2 3
treated by an assignment algorithm where the ob-
Capacity 1 2 1
jective is to assign one control to each area at
minimum total utility cost. While solution of this Mean Error
problem is possible by the Hungarian method, the Score 5 2 4
special structure of the problem (each entry in Frequency
the tableau is a product of frequency times mean Controls of Use
a 1 5 2 4
accuracy score) indicates that the product method b 2 10 4 8
is applicable (Huebner and Ryack, 1961). In this C 3 15 6 12
method, the controls are arranged in order of de- d 4 20 8 16

Downloaded from hfs.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on April 27, 2016
L O U l S E. F R E U N D A N D T H O M A S L. S A D O S K Y August, 1967-297

TABLE
3 TABLE
5
Final Tableau for Transportation Algorithm Final Tableau for Capacitated Transportation
Algorithm
Areos 1 2 3
Copocity I 2 1 Area
1 2 3
M e m Error Copcity
Score 5 2 4 4 2 3
- ~

Frequency Mean Error Score


Controls of Use 5 2 4
a 1 o* 5 1
b 2 0 2 o* Frequency Number to be
C 3 1 0’ 0 Controls of Use Assigned
d 4 4 o+ 2 a 1 A a11 14 12s
b 2 3 iP 0 01
optimal assignments C 3 2 3 w 2
mincost = 58
Table 3, where an o p t h l assignment is indicated
by starred entries. Control a is assigned to area 1, area 1, 1 control b to area 3, and 2 control c’s to
control b to area 3, and controls c and d to area 2. area 2. We assume that the frequency of use of
As a further extension, we allow the number of each of the controls of the ith type is equal (if
each type of control to be assigned to vary, and not, we would subdivide the ith type of control
add additional capacity restrictions relating to into equivalent frequency subclasses).
placing certain controls in certain areas, as shown
in Table 4. The area capacities and mean accuracy Instrument Arrangement
scores are also s h o w in Table 4. We can now
The problem of the optimal arrangement of n
TABU4 instruments in a fixed pattern on a control panel
Initial Tableau for Capacitated Transportation can also be treated as an assignment problem. In
Algorithm the discussion below, we assume that the panel
Area positions available for instrument assignment are
1 2 3 known and the instruments to be assigned are of
Copcity equal size. We also continue to work with hypo-
4 2 3 thetical data.
Mean Error Score Consider the problem of a straight line arrange-
5 2 4 ment of four instruments on a control panel. The
panel could be represented as in Figure 2. Using
~

Frequency Number to be
Controls of Use Assigned the “utility cost” concept, we could measure the
a 1 4 ‘5 ‘ 2 a4 distance component of cost of each position in
b 2 3 I0 4 8
C 3 2 IS 5 12 two ways (See Figure 2): 1) as a function of the
distance, in inches, from the geometric center of
the panel to the center of each instrument (where
formulate the problem as a capacitated trans- the panel center is assumed to be the point of
portation problem, where agah the cell entries normal eye fixation). This approach, in effect,
in the matrix represent the cost of assigning con- implies a larger probability of instrument
trol i to area j. For example, the problem is con- misinterpretation as the distance from the panel
strained so as to aUow placement of no more than center increases; and 2) as a function of the sum of
two control type a’s to areas 1,2, or 3 in the opti- the distances from a particular position to every
mal solution; these restrictions are represented by other position on the panel. In both cases the
numbers in the upper left comer of the appropri- distance component of cost will be termed the
ate cells. The final tableau k shown in Table 5 and “distance value.”
the optimal solution is to assign 2 control a’s to Approach Using Cost Measure I . Using the
area 1, 2 control a’s to area 3, 2 control b’s to Panel center to instrument center as the distance

Downloaded from hfs.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on April 27, 2016
298-AugustJ 1967 HUMAN FACTORS

DIMENSIONS FOR COST MEASURE I arrange the positions in order of increasing dis-
tance value. We would then obtain a solutioncom-
parable to that shown in Table 7 by making as-
signments along the principle diagonal.
Approach Using Cost Measure 2. Here, the dis-
tance value of any position is the sum of the dis-
tances from that position to every other position
on the panel. For example, in Figure 2 the distance
value of positions 1 and 4 would be 2 + 4 + 6 =
12 inches each, and the distance value of positions
2 and 3 would be 2 4 + +
2 = 8 inches each.
With this structure, it is possible toconsider the
probability of transition between instruments a
DIMENSIONS FOR C O S T MEASURE 2 and b as a contributing factor to the optimal
Figure 2. Linear arrangement with distance mea- arrangement. This consideration is not possible
sured between positions and from panel center. under the approach using cost measure 1 above.
The example presented will utilize the probability
of transition matrix shown in Table 8. Structured
value, it is additionally necessary to have data as an assignment problem, the initial tableau
describing the frequency of use of each instru- would appear as in Table 9, where the entry in
ment. Assume the data is as shown in Table 6. the i, jth cell of the matrix represents the dis-
The assignment matrix is also given in Table 6, tance value of the position j times the probabil-
where the cell entries in the matrix represent the ity of transition from instrument i to any other
cost of placing instrument i in position j (i.e., instrument in the panel. For example, this
distance value [from Figure 21 times the frequency probability is shown in Table 8 as the transition
of use per hour). Using the Hungarian method of probability subtotal, Placing instrument 1 in

-
solution, we can derive the final tableau given in position 1 would have an associated cost of .4 X
Table 7. The optimal assignment is given by the 12 4.8. This value appears in the 1.1 cell in the
starred entries in the table. The minimum utility matrix in Table 9. Again, using the Ifungarian
+ + +
cost is then 48 25 19 39 = 131. The op- method of solution, we can derive a final tablcclu
timum assignment is Instrument 1 to position 2, which is presented in Table 10. The minimum
Instrument 2 to position 4, Instrument 3 to posi- utility cost is then 4.8 8 7.2 10.8 30.8.
tion I , and Instrument 4 to position 3. An optimum assignment is Instrument 1 to posi-
+ + + -
This solution could also be obtained using the tion 1, Instrument 2 to position 3, Instrument 3 to
product method. First we would arrange the position 2, and Instrument 4 to position 4.
instruments in descending order according to
decreasing frequency of use. We would next
TABLE 7
TABLE 6 Final Tableau for Assignment Algorithm
Initial Tableau for Assignment Algorithm Position Number
~

Position No. 1 2 3 4
1 2 3 4 Distance Measure
Distance Value 3 1 x 3
3 1 x 3 Instrument Frquerrcy of
Instrument Frequency of .NO.
.-. Use/ l i r .
No. usP/lIT. 1 25 18 0' 0 18
1 25 75 25 25 75 2 13 0 6 6 0 .
2 13 39 13 13 39 3 16 0 ' 0 0 0
3 16 48 16 16 48 4 19 6 0 0 . 6
4 19 57 19 19 57 Optimal Assignmcnts

Downloaded from hfs.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on April 27, 2016
LOUIS E. F R E U N D A N D T H O M A S L . S A D O S K Y August, 1967-299

TABLE 8
Transition Probability Matrix
Transition A way
To Instrument No. Probability From
1 2 3 4 SubToral Panel Total
1 0 .1 0 .3 .4 .6 1 .o
From Instrument No. 2 .2 0 .6 .1 .9 .1 1.o
3 .4 .5 0 .1 1 .o 0 1 .o
4 .2 .3 .4 0 .9 .1 1.o

OPTIMIZATION BASED ON MINIMUM


TABLE
9 EYE TRAVEL DISTANCE
Initial Tableau for Assignment Algorithm
Position Number It is noted that an optimal panel design might
I 2 3 4
be one which minimizes the operator’s eye travel
Distance Factor in using the panel. The objective function will take
12 8 8 12
the form:
Transition
Instrument Probability M h z = Zxij pij
Number Sub-Total
1 .4 4.8 3.2 3.2 4.8 where X i j is the distance from instrument i to
2 .9 10.8 7.2 7.2 10.8 instrumentj and pij is the probability of transition
3 1 12 8 8 12
4 .9 10.8 7.2 7.2 10.8 from instrument i to instrument j. This problem
was formulated for simplex solution. The panel
design considered was a square four-instrument
arrangement shown in Figure 3. The probability
TABLE 10
Final Tableau for Assignment Algorithm
Position Number
I 2 3 4
Distance Factor
12 8 8 12
Transition
Insrrunient Probability
Number Sub-Total I
1
2
3
.4
.9
1
0,
0
.4
2.0
0
0,
2.0
o * o
0
0
.4
I
!
2
I
4 .9 o o o o *
* Optimal assignments

Summary of Utility Cost Method


- 2 4
From the above examples, it is apparent that
Figure 3. Square arrangement of four Instruments.
when a utility cost method is used, the problem of
instrument panel design or workplace layout can
be reduced to a transportation or assignment
problem which can be solved by standard
methods. While rather simple designs have been
used in the above examples, the methods described
can easily be extended to treat more realistic
situations.

Downloaded from hfs.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on April 27, 2016
3OO-August, 1967 HUMAN FACTORS

which, with the recognition that xij = xji, can example problem the optimal arrangement of the
be reduced to: four instruments is shown in Figure 3 by the
Minz = . 3 x ~ .4x13 + + .5X14 + 1 . 1 ~ 2 3+ .4x24 numbers 1 through 4. In our investigation of the
+sx34 problem of arranging 4 instruments in a linear
fashion (as shown in Figure 2) utilizing the sim-
subject to the following restrictions: plex algorithm, we were unable to establish any
systematic procedure for specifying constraint
Xjj 22 i <j inequalities for a feasible solution.
i 1,2,3
=
xij 5 2.8284 j = 2,3,4
Xi2 + + + x23 + x24 +
Xi3 Xi4 x34 = 4(2) +2 SUMMARY
(2.8284) = 13.6568 The following statements seem justified as a
Xi2 + Xi3 5 4.8284 result of this analysis:
Xi3 + Xi4 5 4.8284 1. The workplace layout or control panel
Xi2 4- Xi4 5 4.8284 optimization problem can be approached using
X23 + X24 5 4.8284 the utility cost method described. I t has been
Xz4 4-X34 54.8284 demonstrated that this approach will lend itself
Xi2 4-X23 5 4.8284 to many problems of this type enabling optimi-
X i 2 4- X24 5 4.8284 zation by the product method, the Hungarian
Xi3 + X23 5 4.8284 method, or the assignment or capacitated trans-
X i 3 + X34 5 4.8284 portation algorithms, where the algorithm applied
xi4 + &4 5 4.8284 depends upon the problem structure.
xi4 + x34 5 4.8284 2. The objective of minimization of total eye
x23 + xat S 4.8284 travel distance requires a complex system of con-
straint equations which appear to limit its useful-
The first pair of restrictions set upper and lower ness as a practical technique.
bounds on values of x i j to maintain feasibility.
The third restriction also is included to maintain
feasibility of the solution by restricting the sum of ACKNOWLEDGEMENTS
the possible distances. In this problem, we are
essentially asking which instruments should be The authors wish to express thanks to Pro-
“caddycorner” from each other. The remaining fessors R.M. Thrall of the Department of Indus-
feasibility problem is to define which instruments trial Engineering and R.W.Pew of the Human
can be across the diagonals from each other in a performance Center at the University of Mich-
feasible solution. This is done by noting that the igan for guidance on this project.
sum of pairs of distances which have a common
digit in their subscript must be either 4.0 (both REFERENCES
distances are along the rim), or 4.8284 (one dis-
Dantzig, G. B. Litiear programming and exlensiolu.
tance is along the rim and the other distance is Pnnceton, N.J.: Princeton University Press, 1963.
across a diagonal). These restrictions are enumer- Deiinger, R. L. Rocess sampling. workplace ar-
ated in the last group of constraints given above. rangements and operator activity levels. USAF-
Even though the number of instruments was WADC, Engineering Wchology Branch Unpub
S m a l l s the problem demanded constraints which lished Manuscript, 1958.
Fitts, M, A study oflocation discriminationability.
it large for hand computation. Thus, a In P. M. Fitts (Ed.) Psychologkd researrh on equip
Computer programmed simplex routine was ment design. Washington, D.C. : U.S. Government

we can then construct a suitable panel. For this TR 61-143,1961.

Downloaded from hfs.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on April 27, 2016

You might also like