COIPYI. & 0~s Res. Vol. 6, pp.

147-N
Pergamon Press Ltd.. 1979. Printed in Great Britain zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

A MULTI-MACHINE LABOR ASSIGNMENT FOR

VARIABLE OPERATOR SERVICE TIMES

RAIESHDUBE*and E. A. ELSAYED~
College of Engineering, Rutgers University, Piscataway, NJ 08854, U.S.A.

Senpe and purpose--The determination of operator work loads in multi-machine labor assignment problems
where machine interference prevails is a common problem in industry. This paper presents a heuristic
algorithm that provides optimum or near-optimum solutions for the labor assignment problem where the
total cost of idle times for both labor and machines is minimized. A computer program is developed to
implement the assignment algorithm for large size problems.

Abstract-This paper presents an approach to the multi-machine labor assignment problem, where a single
operator can serve a number of semi-automatic machines simultaneously in a synchronous service
discipline. The operator service times are assumed to be normally distributed, while the machine processing
times are deterministic. This paper provides optimum or near-optimum solutions for the assignment at a
minimum combined cost of idle times for both labor and machines. Feasibility checks are incorporated in
the solution procedure to meet the due date and maximum operator travel distance restrictions. A computer
program is developed to implement the assignment procedure for large size problems.

INTRODUCTION

Man and semi-automatic machine relationships are usually one of three types: (i) synchronous
or cyclic servicing, where both the operators and machines are occupied the whole cycle, (ii)
random servicing, and (iii) a combination of (i) and (ii).
It is pertinent in any production operation to find an optimum assignment of men to
machines. This assignment problem has been dealt with in two distinct approaches (Miller and
Berry [ 1I). The first approach deals with an assignment problem as a queueing model, where the
operator’s service time and machine processing time are stochastic in nature. This approach is
useful for production systems where minimal information is available regarding the jobs to be
processed on individual machines. An account of these models can be found in Ashcroft [2] and
Fetter[3].
The second approach is concerned with the problem of preparing machine assignments for
operators after individual jobs (orders) have been assigned to machines for processing. The
operator’s service time and machine processing time are known and deterministic in nature. The
work of Burgess[4], Conway et al.[S], Jones[6], Killingback[7] and Miller and Berry[l, 81 are
oriented in this direction.
This paper presents a modified algorithm for the Labor Assignment Heuristic developed
earlier by Miller and Berry[8]. This algorithm provides optimum or near-optimum solutions to
the labor assignment problem. It determines the number of operators (N) required to serve (M)
machines under consideration; and the individual machine assignments for the operators in
order to minimize the combined operator and machine idle time costs. The operators’ service
times are assumed to be independent and normally distributed while machine processing times
are deterministic in nature. A computer program is developed to find the optimum solution for
the assignment problem.

ANALYSIS OFTHE SYSTEM

The following assumptions will be considered in describing and analyzing the algorithm:
(1) There is no restriction on the number of machines that an operator can operate within
the distance specified in (2);
*Rajesh Dube is a graduate student and Research Assistant in Industrial Engineering and Operations Research at
Rutgers University. His research interests are in the areas of production systems, cluster analysis and information systems.
He holdsa B. E. in Mechanical Engineering from lndore University, India, and an M. Tech. in Industrial and Management
Engineering from the Indian Institute of Technology, Kanpur.
tE. A. Elsayed is currently an Assistant Professor in the Department of Mechanical, Industrial and Aerospace
Engineering at Rutgers University. His research interests lie mainly in the area of stochastic processes and production systems.
He received his Ph.D. in Industrial Engineering from the University of Windsor, Canada, and he has taught previously at the
University of Utah and Cairo University, Egypt.

147
148 RMESH DUBEand E. A. ELSAYED

(2) The distance between any two machines assigned to a particular operator is less than a
pre-specified distance called the “Operator Maximum Travel Distance”. Due to this restriction,
the travel time between machines assigned to an operator is considered negligible;
(3) A minimum acceptable production rate (Z?i)can be stated for each job (machine) so that
the scheduled delivery date can be met;
(4) Machine processing times (Pi) are deterministic in nature;
(5) While they cannot be segmented, the operator service times are assumed to be normally
distributed with parameters pi and a/. This allows variations in the operators’ service times and
their productivity on different machines.
(6) The idle labor cost (ILL’) is identical for all men, the value of ZLC can be estimated as
the averaged hourly wage rate for the shop; and
(7) Machines are serviced according to synchronous (cyclic) discipline.
The assumptions above, with the exceptions of (I), (2) and (5) are the same as those stated
by Miller and Berry[8]. The approach presented here is more realistic with the inchtsion of the
due date and operator travel distance restrictions as well as the variability in operator service
times.
The first step towards the solution of the labor assignment problem is to develop expres-
sions ‘for finding the costs of idle labor and machine times. The labor assignments are
determined with the objective of minimizing the combined cost of these times. Costs associated
with the assignment problem can be determined as follows:
(1) The “effective machine cycle time” (NT): which is the time between successive visits
by the jth operator to each machine assigned to him.

where S, the set of all machines assigned to operator j; pi, mean operator service time of ith
machine (in minutes); ut, variance in the mean service time of the ith machine; Pi, power feed
time (processing time) of the ith job/machine (in minutes); a, probability factor associated with
the level of confidence desired in the assignment.
(2) Idle labor cost per shift (I&) for the jth operator is computed as follows:

(2)

where t, working time/shift (usually 480 min per shift); C,, cost of idle labor ($/hr).
(3) The idle machine time cost per work shift zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPO
(ZMCJ is computed for the jth operator as
follows:

where r+ number of machines in the set (4); Cz, cost of idle machine ($/hr).
The total idle time cost (TZCj) for the jth operator per shift is:

TZCj= ILCj + IMCj,: (4)

The total cost of combined idle machine and labor time per shift for all operators (TIC) is,

where N is the number of operators required to run the M machines and I I N I M.

The modified heuristic starts with an initial assignment in which a single operator is assigned
to each machine (N = M). In this case, 5 IMCj = 0, while 5 ILCj is the only idle cost zyxwvutsr
j=l j=l
 A multi-machine labor assignment for variable operator service times 149

involved. The heuristic then systematically searches for an improved solution by proceeding
from point (a) to point (b), i.e. reducing the number of operators, as shown in Fig. 1 where m
is evaluated and compared with m at&t (a), if TIC at zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLK
b < 8? at a, the process is repeated
until we have a minimum value for TIC at point (c). This procedure is the same as that
developed by Miller and Berry[8].
The algorithm can be summarized in the following steps:
(1) Start with (N = M) operators and assign one machine to each operator. If S denotes the
set of machines.assigned to the jth operator, initialize the sets as follows:

Si=(i}forVj= l(l)N.

(2) Determine the maximum allowable cycle time (Xi) for each job (machine) so as to meet
the due date restrictions.
Xi = 9 minutes for Vi = l( I)M
I

where 9, number of shifts remaining till due date for job (i); ti, working minutes per shift; Qi,
order quantity for job (i).
(3) Select the maximum travel distance (0) between any two machines assigned to the same
operator, and the value of the constant (a) depending on the level of confidence desired in the
solution.
(4) Generate the sets:

for Vj = 1 (1) N and Vk = j(l)N.

600.

100

0 I
I 2 3 4 5 6 7 e 9 IO

Number of Machines

Fig. I. Behavior of objective function for a IO-machine problem at 50% confidence level.
 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
R.MESHDUBE and E. A. ELSAYED
IS0 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

Compute the cycle times (MC&) using equation (1). MCqk represents the resulting cycle
time if the machines belonging to sets Sj and S, are assigned to a single operator.
(5) Check whether MCTjk 5 .& for Vi E Sjk if MC’7jk> & for some i E Sjk; set T1& = W,
(where W is a very large number), then go to Step 8.
(6) Check whether dir s 6 for Vi, zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIH
1 E Sjk. If dir is more than the maximum travel distance
(e), then set TI& = W and go to Step 8. It should be noted that di, is the distance between
machines (i) and (I).
(7) Calculate KC and KMC,I,using equations (2) and (3). Consequently, the combined cost
of idle machine and labor times is computed using equation (4):

Elements of TZC matrix represent the total idle time cost if the machines belonging to the sets
S and Sk are combined and assigned to one operator. The total cost associated with the current
solution can be found by summing the diagonal elements Cjj of the TIC matrix. These elements
(Cjj) represent the idle time cost associated with each operator (j):

TIc=$J TIci.
j=i

(8) Find the matrix of cost savings (2,) ass~iated with each solution combining (4) and
(Sk), whose elements are given by:

&=TIcj+TK’~--T&; j=I(l)N,k=(j+I)(I)N and j+k.

If all (& SO), the current solution is optimal. If any (Zjk >O); then find

Z* = max {,ZJ j = l(i)(W - I); k = (j t I)(l)N

Let Z,, = 2* such that (r < s).

(9) Assign machines belonging to operators (I) and (s) to operator (r), i.e. combine sets S,
and S, to form a new set S,.

s, = {Sr u SJ.

Since operator(s) has been eliminated, renumber the operators (S + I) to(N) with an increment of
I, i.e.
s, = S/,l,VS( I)(N - I).

(10) Let N = N - I and go to Step (4).

A flow diagram for the above algorithm is given in Fig. 2. Also, a computer program was
developed to utilize such an algorithm for different levels of confidence. The input data to the
program is as follows:
(I) Idle labor cost Ci in $Ihr.
(2) Machine idle time cost CL,$Ihr.
(3) Maximum operator travel distance (@).
(4) Number of minutes per shift (t).
(5) Machine numbers and their processing times (pi).
(6) Operators’ service time (hi) and corresponding variance (g?)“>.
(7) Order quantity for any job (Q).
(8) Number of shifts remaining till due date for any job (Di).
(9) The matrix of intermachine distances.
(10) The probability constant corresponding to confidence level.
The output of this program contains the following for each confidence level:
(1) Number of machines assigned to every operator;
(2) Which machine is assigned to which operator;
 A multi-machine labor assignment for variable operator service times 151

(3) Cost of initial solution 2;

(4) Cost of final (optimal) solution Z*; and
(5) Operators’ cycle time.
Using the developed computer program for solving a lo-machine labor assignment problem
the following results are obtained: zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGF

A IO-Muchine problem
The details regarding the jobs to be processed are as follows:

Machine Order
no.

I 1001 6 1.0 7 200 20 10.00 48.00

2 1002 2 0.3 13 . 180 11 16.36 29.33
3 1003 7 1.5 4 75 17 4.41 109.80
4 1004 3 0.5 8 300 16 18.75 25.60
5 I005 3 0.5 7 79 5 15.80 30.38
6 1006 5 1.2 9 85 5 17.00 28.24
7 1007 4 0.5 6 115 4 28.75 16.70
8 1008 6 1.5 2 210 8 26.25 18.29
9 1009 8 1.6 7 211 9 23.44 20.47
10 1010 12 2.1 4 31 2 15.50 30.97

Ri = Minimum acceptable production rate = QjlJ units/shift.

Xi = Maximum acceptable cycle time = zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONM
r,lRi minutes.

The matrix of inter-machine distance is as follows:

1 2 3 4 5 6 7 8 9 IO

7 13 10 9 II 4 9 7 11
6 4 3 9 11 3 5 8
16 14 2 9 II IO 8
15 15 12 7 6 15
12 10 17 9 6
II 18 13 3
IO 7 9
12 4
11
Idle labor cost (C,) = $5.OO/hr.
Idle machine time cost (Cr) = SlO.Whr.
Maximum operator travel distance (8) = 15.00 ft.
Number of minutes per shift = 480 min.

The following results were obtained:

Case OP, OPZ OP, OP4 OP, OP6 % Confidence

(1)
(I%, 2,976
(15.00) (8.k)
216.22 76.49 50.00

(2) 4,537 (lE2) 208.49 83.25 69.15

(ltd.sbs, (lE3, (11161) (11.35) (Sk)
(3) 1,2,6 4,577 - 201.34 85.02 84.13
(15.55) (12796, (11.71) (9.822,
(4) 1,2,6 4,597 8 (ld.& 194.71 85.68 93.32
(15.82) (12384) (12.06) (9.84)
(5) 1,296 4,577 R (GO, 188.53 85.97 97.72
(16.95) (13345) (12.45) (li45) (17.53)
(6) 1,2,6 498 10 182.77 92.37 99.374
(16.95) (14306, (12.77) (lY7, (18916,
(7) 1.2.6 (‘Yii2) 121.69 97.70 99.861
(17.74) (14367) $4, (:;.I, (18979) (20.35)

*Job number
tCycle time
152 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQ
Rum BABE and E. A. ELSAYED

04 = Operator number 0,
Zao= cost of initial solution, at specified confidence level,
22 = cost of final solution, at specified confidence level.
It can be observed that the cost of idle time for the initial solution decreases with the
increasing level of confidence. This is due to the fact that the assumption of higher confidence
level leads to increased values of service time hence cycle time, which is the sum of the service
and process time, also increases. This resluts in a lower number of cycles per shift and thereby
in a lower cost of the initial assignments.
The behavior of the objective function was studied for both unconstrained (distances and
due date constraints are inactive) and constrained probfems. Figure 1 shows the behavior of the

FIND MINIMUM
ACCEPTAGLE
EDUCE
RATE (I?,) FOR
EACH Jo8

I
FIND MINIMUM
ACCEPTABLE
CYCLE TIME
(X,) FOR EACH
JOB

1
PRINT INPUT
DATA
REGARDING THE

7
INITIALIZE
WI)
V,=l(l)N

i 1

Fig. 2.
 A zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
multi-machine labor assignment for variable operator service times 153

c zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

YES

N +N -1 flND
nc,

Fig. 2. Flow diagram for the assignment problem.

objective function for a 10 machine problem. It is obvious that the number of operators for an
unconstrained problem can be reduced to one, whereas, in a constrained problem it cannot be
reduced below a certain number due to the imposed constraints.
The program thus successfully analyzes the multi-machine labor assignment problem and
provides a variety of solutions with different levels of confidence.

COMPUTATIONAL EXPERIENCE
The computer program developed in this paper was written in FORTRAN IV and run on an
IBM 370/168 computer. The program was executed for 6 randomly generated assignment
154 RAJESHDutm and E, A. ELSAYED

SOK

40K

IOK

5 IO 15 20
Number of Machines

Fig. 3. Number of machines vs computation time and core requirements.

problems. The computation times and core requ~ements for these problems are shown in Fig.
3. It should be noted that the computer time used by the program is negligibleas compared to
the savings that can be realized for larger problems.
On comparing the algorithm with one obvious method of solving the assignment problem,
namely; the man-loading heuristic developed by Miller and Berry[8], it was found that the
proposed algorithm gives a lower idle time cost in most cases at the expense of higher
computational time.
CONCLUSIONS

The approach presented here provides an optimal solution for the multi-machine labor
assignment problem in production situations, where semi-automatic machines are used. The
developed computer program can be easily implemented in these sit~tions to make day to day
assign~nts and thus movement can select a solution having an e~ono~~aliy justified level
of confidence.
The distance matrix can be stored on the computer itself and other time values, if possible,
can also be stored in terms of standard times. This will automate the procedure further and the
task‘will become very simple. Only the necessary values need to be recovered from the data
files.

Ackrrow/edgemenrs-The authors would like to thank the Editor. and the referees for their helpful suggestionsand
comments.

REFERENCES
1. J. G. M&r and W. L. Berry, The assignmentof men to machines:an applicationof branchand bound.&&ion .%i. 8(l),
56-72 fl977).
zyxwvutsrqponmlkjihgfedcbaZYXWV
2. H. Ashcroft, The pr~uctivity of several machinesunder the care of one operator. R. Stat. Sm. L 12, 145-151 (1950).
R. N. Fetter, The assignmentof operatorsto service ~mi-au~matic machines.J. ~ndu~t~u~Engng 6(S), 22-30 ( 195.5).
:: A. R. Burgess,Commentson simulationin the applicationof wage incentivesto multiple machines.J. industriul Engng
13(4), 264 (1%2).
5. R. W. Conway, W. L. Maxwell and J-J.W. Saimpson,Gn the cyclic servicingof semi-automaticmachines.J. industtia/
Engng U(2), 105-10s (1%2).
6. D. Jones, Graphical determination of work loads for multiple machine assignments.J. fndustriol Engng q2), 16-25
(1953).
7. I. Killingback, Cyclic interferencebetween two machineson differentwork. ht. J. ProductionRes. 3(2), 115-120(1964).
8. I. G. Miller and W. L. Berry, Heuristic methodsfor assigningmen to machines:An experimentalanalysis. AIIE Trans.
q2), 97-104 (1974).
9. G. Clarke and J. W. Wright, Schedulingof vehicles from a central agent to a number of delivery points. Ops Res. It,
568-581 (1964).