international journal of

production
economics

ELSEVIER

Int. J. Production

Simulation

analysis

Economics

of a pull-push system for an electronic

assembly line
Mehmet

Industrial and $vstems Engineering,

51 (1997) 205-214

Savsar*

Department of Mechanical and Industrial Engineering,

Kuwait University, P.O. Box 5969, &fat, Kuwait
Received

11 July 1996; accepted

7 February

College of Engineering

and Petroleum.

1997

Abstract

This study deals with simulation modeling and analysis of an electronic assembly line. Weekly scheduled demands for
printed circuit boards (PCBs) are met from its Final Board Inventory (FBI). The demand triggers production at the first
station by signaling for replenishment of a quantity equal to the demand. The assembly operations themselves are
performed using a push system of production control. The objective of the simulation analysis is to determine the
minimum number of batches in the system, including the Work-In-Process (WIP) and Finished Board Inventory, to meet
a certain percentage of demand on time. The assembly line has enough capacity to meet the annual demand, but random
fluctuations in assembly operations disturb the weekly production schedule. Therefore, the aim is to meet weekly
scheduled demand on time to avoid additional cost of reshipments.
Keywords: Just-in-time;

Kanbans;

Pull-push system; Simulation; Work-in-process

1. Introduction

Production
and assembly
systems which are
driven directly by the demand for finished products
attempt to minimize product shortages as well as
excess inventory.
These systems have attracted
interest in recent years particularly
in the case of
high-volume
assembly systems. An effective production control system is the one that produces
the right amount
of parts, at the right time, at
a competitive cost. Introduced
by Toyota in Japan,

*Tel.: 966-l-4645490;
fax: 966-l-4676652;
e-mail: F45MOO9@KSU.EDU.SA.
0925.5273/97/$17.00
Copyright
PII SO925-5273(97)00055-8

a production
control technique called just-in-time
(JIT) or pull system of production
has been implemented by many companies around the world.
A common system is based on information
carrying
cards called kanbans
which are attached to the
products or containers to initiate production
which
is synchronized
with demand. Kanban-controlled
systems using cards, containers,
balls, space outlines, etc. have been introduced
to many types of
industries
including
the electronic assembly lines
discussed in this paper. Kanban
implementation
can be full or partial for the line.
Several studies have been carried out on the
implementation
and efficiency of JIT systems in
different industries under different operational
conditions. Huang et al. (1983) present a simulation

% 1997 Elsevier Science B.V. All rights reserved

206

M. Savsarllnt.

J. Production

analysis of a JIT system for a four-stage assembly

line. They investigate the effects of variable processing time, variable master production
schedule, and
imbalances between the stations. Ebrahimpour
and
Fathi (1984) studied the behavior
of inventory
levels in a kanban controlled
production
system
under
cyclical
and steadily
growing
demand.
Fallow and Browne (1988) present a simulation
study for a JIT system. Sarker and Fitzsimmons
(1989) compare the performance
of pull systems
with that of push systems under different operational conditions.
Yamashita
et al. (1989) present
a theoretical model for a production
line driven by
demand. They determine the line efficiency, which
is measured
in terms of average shortages
and
average in-process
inventory
levels. Olhager and
Ostlund (1990) present the integration
of push and
pull systems and its benefits in a manufacturing
environment.
Wang
and Wang (1990) present
a Markov
model to approximate
the optimum
number of kanbans for reducing in-process inventory. One of the most comprehensive
literature
review on JIT systems is presented by Golhar and
Stamm (1991) who identify and classify 860 articles
in the area.
Bard and Golany (1991) developed a planning
model to assist managers in determining
optimal
kanban
policies at each station
of a JIT line.
Hodgson and Wang (1991) present Markov decision process models for hybrid push/pull
control
strategies in multistage systems. Golhar and Sarker
(1992) develop a model for economic manufacturing quantity in a JIT delivery system. Chu and Shih
(1992) summarize
and classify the simulation
studies performed on JIT systems. Berkley (1993)
presents a simulation
model and its results for
determining
minimum performance
levels for twocard kanban controlled lines. Fukukawa and Hong
(1993) present a mixed integer goal programming
model to determine the optimal number of kanbans
in a JIT production
system and Zhuang
(1994)
studied the optimum
levels of price and delivery
frequency between a supplier and a JIT buyer firm.
Countless
number of other JIT applications
and
related models can be seen in the literature. Most
of the literature
deal with the efficiency of JIT
systems under different
operational
conditions.
Either mathematical
models are developed based

Economics

51 (1997) 205-214

on restrictive assumptions
or simulation
models
are utilized in the analysis of JIT systems. Some
companies also use trial and error procedures such
as reducing the number of Kanbans
until the system comes to a halt, i.e., gets into trouble.
This paper discusses simulation
analysis of an
actual electronic assembly line which was designed
to produce a printed circuit board (PCB) in mass
quantities
to be used in communication
systems.
The assembly line under consideration
was newly
designed to meet the demand for the PCBs for
a period of five years. The annual demand for the
PCBs has been predetermined
since the company
had a contract to produce certain number of PCBs
per year. A weekly production
rate of 670 PCBs is
required to meet the demand. Since the demand is
always met in batches of 10 boards, 670 PCBs
represent 67 batches. The PCBs assembled during
one week are transferred at the end of the week to
another line where these boards are integrated into
communication
equipment. Fig. 1 shows the layout
of the assembly line specifically designed for the
required PCB. Initial assembly operation
at the
first stage is triggered by the demand from the
Finished
Board Inventory
(FBI). Thus the line
starts operation
according
to the demand
of
finished boards. After the assembly operation
is
started at the first station, the assembled batches of
boards are pushed down the line until they can be
put in the Finished Board Inventory
in batches of
10 units.
The main objective of the study was to determine
the minimum
number
of kanbans
attached
to
batches of PCBs to meet a certain percentage of
demand on time. Although demand could be met
by the line capacity, meeting the demand on time
was the main target. Shipments were made at the
end of each week and any units of demand not met
at the time of shipment would require additional
shipment cost. Management
wanted to determine
the minimum
number of kanbans
needed, minimum intermediate
buffer levels and the minimum
FBI level realized if the specified percentage
of
demand was to be met on time. Daily demand was
about 13 batches (67 batches per week/5 days =
13.4).
The simulation
model and analysis presented in
this study is different from the previous studies

M. Savsar/Int.

Kanban

SMT

Release
-

Stations

Heat
Teat

J. Production

WPl

WP4

Economics

Pre Assy &

Manual Ins.

Final
Inspection

51 (1997) 205-214

Wave
Solder

WP3

Zentil
Test

207

Inspection
&Touch Up

WP2

-I

Post
Wave

I
1

Rework

Demand

Fig. 1. Layout

of the electronic

assembly

line simulated

in several ways. First, it addresses a real case application in which a combination

of pull and push
systems of production
control is implemented
in
a factory to operate the system based on known
demand. The kanbans from the last station on the
line are transferred
to the first station weekly (or
daily if weekly demand is not met on time) instead
of the immediately preceding station as practiced in
most actual kanban
implementations.
Second, it
answers a typical question related to the effects of
the number
of kanbans
and
the minimum
WIP/FBI
levels on percentage of demand met on
time; and finally, it analyzes the fluctuation
of FBI
level with respect to time.

2. Line operation
The assembly line under consideration
had nine
stations as shown in Fig. 1. A combination
of pull/
push strategy was used to control product flow and
operation
of the line. Finished
PCB assemblies,
which are in batches of 10 units, are pulled from the
output of the last station and an equivalent number

(dark circles indicate

work-in-process

buffers_WIP)

of kanbans are transferred to the input buffer of the

first station to trigger the production
which is to
replenish demand. From this point onward, a push
system of control is used. The batch of assemblies
are pushed down the line along the consecutive
stations as far as they can go without any signal
from a succeeding station to its preceding station
that would be practiced in a pull system of production control. Basic steps of line operation
are
described as follows:
1. Orders
of 67 batches
of Printed
Circuit
Boards (PCBs) are received every week from a
Communication
Equipment
Assembly Line. Each
batch consists of 10 PCBs.
2. As the orders are received at fixed time intervals (at the beginning
of each week) and in fixed
order quantities (67 batches), they are derived from
the finished board inventory
(FBI) depending
on
the availability
of FBs in inventory. If there are 67
or more batches of PCBs in the FBI, the demand is
met completely; on the other hand, if there are less
than 67 batches of PCBs in the FB inventory, the
demand is met partially depending on the availability in the FBI.

208

M. Savsnrllnt.

J. Production

3. As the demand is met from the FBI in batches,

the same number of production
ordering kanbans
are released to the first assembly work center. This
work center consists of a rigidly linked series of
automated
stations
called the Surface Mount
Technology
Stations (SMT). The following notation is used for the demand which is met on time.
Let,
Di = demand in batches at the end of the ith week,
Fi = number of batches of PCBs available in the
FBI at the end of the ith week,
Mi = demand met (in batches) on time at the end of
the ith week,
Pi = percentage of demand met on time at the end
of the ith week.
Thus, the demand met on time in the ith week, hiii,
is found in simulation
as follows:
Mi=Fi

ifFi<Di,

Mi=Di

ifFi>Di

Pi = lOO(Mi/Di)%
when
Fi < Di
lOO(Di/Di)% = 100% when Fi > Di

and

Pi =

One kanban is attached to each tray that carries

a batch of 10 boards. The total number of kanbans
(number of empty lo-board
trays) released to the
first assembly station, i.e., the production
order
size, is always equal to Mi, which changes according to the demand met. The SMT (Surface Mount
Technology)
stations are rigidly linked automated
assembly machines which place and solder the electronic components
on the PCB at specified locations by using advanced
assembly and soldering
technologies.
The printed circuit board is screen
printed first with solder paste; then a chip shooter
places the components
at appropriate
places at an
extremely fast rate. The PCB is then transferred
into an oven to melt the solder paste that solders
contacts on the electronic parts on the PCB.
4. After the SMT, the PCBs are placed in
a Work-In-Process
(WIP) buffer (WPI in Fig. 1) to
be transferred to the next station, which is the Pre
Assembly and Manual Insertion
station. At this
work center, through hole components
are manually placed into the designated holes.
5. The PCBs are next transferred
to a Wave
Solder Machine where the boards pass through
a solder bath for soldering
the through
hole
components
to the PCB.

Economics

51 (1997) 205-214

6. Next the boards are transferred to an Inspection and Touch up station where they are visually
inspected and the defective solder joints are resoldered.
7. The boards are transferred to a work station
called Post Wave work station. Here, two work
elements are performed. A face plate is assembled
to the board and a label is pasted onto it. The
boards are placed in a 2nd WIP buffer (WP2).
8. After completion
of the assembly operations,
the PCBs are transferred
to an In-Circuit
Test
Machine
Center
called Zentil.
The same test
machine and the fixtures attached to it are used to
perform two tests on each PCB consecutively.
First,
an in-circuit test is performed on all parts of the
PCB to detect the defective parts. Then, a functional test is performed to see if the board does its
functions properly. PCBs which pass the test are
placed in a 3rd WIP buffer (WP3) after the incircuit test machine. Any board found defective is
routed to a Rework Station where it is repaired and
sent back to the in-circuit test machine. A PCB,
which is found defective in the middle of the test
operation
at Zentil test machine, is immediately
transferred to the rework station without completing the rest of the test.
9. After the Zentil test, the boards are routed
from the Zentil WIP buffer (WP3) to the Final
Inspection
Station where an overall inspection
is
done including the quality, labeling and other features of the board. Any rejected boards are sent to
the Rework Station. After the rework operation
and repair of any defects, the boards are routed
back to the Zentil test station. The PCBs which
pass the final inspection are placed in a final WIP
buffer (WP4).
10. After the final visual inspection,
the boards
are sent to a Heat Test Station. Here, the boards
are loaded onto racks and subjected to operation
under a certain temperature.
The PCBs undergo
this test during the night shift for 8 h. During the
day shift, while the boards arriving to the heat test
are loaded onto racks for the test, those boards
which have completed the test are unloaded from
the racks by an operator. Any board found defective in the heat test is routed to the rework station
for repair. After the repair they are sent back to the
Zentil test machine.

M. Savsar/Int.

J. Production

11. From the heat test, the completed boards are

transferred to the Final Board Inventory
(FBI) to
meet the weekly demand.
Initially the production
manager needed to estimate the minimum
capacity of the FBI and the
WIPs in order to meet a certain percentage of the
demand on time, possibly 100%. This requirement
was related to determining
the minimum number of
kanbans
needed to circulate in the system. Each
kanban was attached to a batch of ten PCBs. The
objective was to determine the minimum number of
kanbans (batches of ten PCBs) allowed at the WIP
and FBI inventories to meet a certain percentage of
demand on time. Of course, the production line was
designed with a capacity such that all the demand
would be met. However, due to random fluctuation
in some station operation
times and random failures in equipment,
the FBI level also fluctuates.
Some of the demand is met on time while some
other is not. The fixed weekly demand is met from
the FB inventory depending on the PCB availability. However,
any number
of delayed boards,
are completed
during
the following
week and
the demand is eventually met but with extra shipment cost. The problem was to maximize the percentage of boards which met the demand on time.
The percentage
of the demand
met on time at
the end of the week was the main performance
measure.
As it is seen from the description of the assembly
line, the line was started according to the demand
for the finished board. As the demand was met from
the FB inventory,
the same number of kanbans
were routed to the first station (SMT) on the line to
trigger the production. This was an implementation
of pull system of production
control with kanbans
where the demand triggers the assembly at the first
station by the last work station, namely the FB
inventory.
After the first assembly station point
onward, the push system controls production.
The
boards are pushed down the line (in batches of 10)
until they reach the FB inventory.

3. Simulation

modeling and analysis

The purpose of this work was to build a simulation model to analyze the assembly line illustrated

Economics

51 (1997) 205-214

209

in Fig. 1. The simulation model is intended to answer several questions related to the assembly line
performance
before its actual operation.
As mentioned in the previous section, the management
was
mainly interested
in determining
the minimum
number of kanbans and the WIP buffer and FBI
capacities that would result in meeting a certain
percentage of demand on time. In addition, production cycle time estimates, production
output rates,
work station utilizations,
and FB inventory
fluctuation levels over time were required. The simulation model of the assembly line was constructed
using the SIMAN language developed by Pegden
(1987). Fig. 2 shows the general block diagram of
the simulation
model.
One of the most important
aspects of the
modeling was the data collection phase. As outlined in Fig. 1, the production
line consisted of
9 work stations as follows:
1. SMT (Surface Mount Technology)
Center,
2. Preassembly
and Manual Insertion Stations,
3. Wave Solder Station,
4. Inspection
and Touch-Up
Station,
5. Post Wave Assembly,
6. Zentil In-Circuit
Test Station,
7. Final Inspection
Station,
8. Heat Test Station,
9. Rework Station.
Assembly operation times and the related distributions were estimated by the Industrial
Engineering
department
in the factory. Table 1 presents the
estimated process times and the related distributions for each station on the line. Some of these
times were obtained from equipment specifications
while most of the others were obtained
from a
similar assembly line operated elsewhere.
After a detailed study of the line, the following
facts were obtained
and assumptions
were made
related to the operational
characteristics
of the
assembly line.
1. Total weekly demand for the PCBs was estimated to be 67 batches, each batch consisting
of
10 boards. This was determined
according to the
annual demand and its uniform distribution
over
the year. The demand,
which was indicated
to
be fixed every week, was generated in groups of
67 batches at the end of each week and was met
from the FB inventory. However, those batches of

210

M. Savsarlht.

J. Production

at Scheduled Times

Determine the Number

of Batches in FBI (Fi)

Transfer Mi Kmbms
to the First Station to

Assemble Withdrawn Quant.

1
Simulate Rest of the Line
using the Push System
1
Store Weeks Assembled
Quantity in the FBI
Fig. 2. General
developed.

block

diagram

of

the

simulation

model

demand that could not be met on time were met as

the boards were assembled during the following
week.
2. The actual working time per day was estimated as 425 min (excluding lunch and the breaks).
Thus the total operation time in a 5-day week was
2125 min and in a 50-week year, it was 106 250 min
(assuming 2 weeks of holiday during the year).

Economics

51 (1997) 205-214

3. All work stations

are expected to operate
425 min per day except the SMT station which was
expected to operate 395 min/day since 30 min/day
was required for set up and cleaning.
4. For the in-circuit test operation,
two Zentil
machines were utilized since one was not enough to
meet the demand. Therefore the capacity of the
in-circuit
test station was doubled
by assuming
two parallel resources (i.e., test machines) in that
station.
5. The heat test was conducted during the night
shift on all the boards simultaneously
loaded onto
the test racks. Therefore, operation rate at the heat
test station was assumed to be the working rate of
the operator at this station during the day shift,
which was 2.72 minIboard
(or 27.2 min/batch)
for
loading and unloading
operation.
6. Using the available data from a similar factory, SMT failures, part jams, reel replacements,
part unavailability,
etc. were all combined into one
failure or stoppage rate for the station. It was estimated to be exponentially
distributed
with a mean
of 60 minutes between the stoppages. The repair
time or the time to clear up the jams and to replace
the part reels was estimated
to be exponentially
distributed
with a mean of 20 min.
7. For the in-circuit test station, time between
failures was also estimated to be exponential
with
a mean of 60 min, while the repair time or the time
to restore the test machine to its working condition
was estimated to be exponential
with a mean of
15 min.
8. From the data available, the percent of boards
accepted at the in-circuit test station was estimated
to be normally distributed with a mean of 70% and
a standard deviation of 7%. Thus, on the average
30% are rejected and sent to the rework station.
Table 2 presents the percent of boards rejected at
other stations. It was 5% at the final inspection
station and 1% at the heat test station. All the
boards rejected at any test station are routed to the
rework station. After rework, all the boards are
transferred to the in-circuit test station regardless
of the station they were sent from.
9. The operation
time at the in-circuit test station was 4 min/board
or 40 min/batch.
However,
those boards which failed the test were removed
immediately
from the station without completing

211

M. Saasar/Int. J. Production Economics 51 (1997) 205-214

Table 1
Assembly

line stations

operation

time distributions

and related

parameters
(Mean,

Std. D.)

Process

SMT
Pre Assmb. and Man. Ins.
Wave Solder
Insp. and Touch Up
Post Wave
In-Circuit Test (Zentil)

Constant
Normal
Constant
Normal
Normal
Constant

1.89 (Minutes/Board)
2.00, 0.20
0.5
3.00, 0.30 (two operators)
1.575, 0.20
4.00 (two parallel m/c)

Forfuiled boards
Final Inspection
Heat Test
Rework Station

Uniform
Normal
Constant
Uniform

2.00, 4.00
3.00, 0.30
2.72
3.00, 5.00

Table 2
Percentage

of rejection

time distribution

Parameters

Station

at test stations

Test station

Percentage

in-Circuit test
Final Inspection
Heat Test

30%
5%
1%

of rejection

rest of the test. The partial test operation time for

the failed parts was estimated to be uniformly distributed between 2 and 4 min.
A brief description of the simulation model of the
electronics assembly line under consideration
is as
follows:
At the end of each week, 67 entities representing
the 67 batches of demand are created and transferred to the FB inventory queue at the end of the line.
From the availability
of the boards in the FBI, the
demand is met at the time of requirement.
If there
are enough batches of boards in the FB inventory
(i.e., Fi >Di), then all demand is met. On the other
hand, if there are less boards in the FBI than the
demand (i.e., Fi <Oi), only the amount available is
shipped to meet the demand partially and the rest
of the demand is met later in the week as the boards
are assembled.
However, meeting the demand at
later time during the next week is undesirable since
it results in additional
shipment costs to the company. A counter is kept in the simulation
program
to determine
the demand met on time and the
demand not met on time.

For each batch of demand met from the FBI,

a card, representing
a kanban, is sent to the first
(SMT) station on the assembly line. For example, if
all the weekly demand is met, 67 cards (kanbans)
are routed to the SMT station. On the other hand,
if less than 67 batches of demand are met, only this
number
of kanban
cards are sent to the SMT
station. Each of the stations are represented
as
resources and the Work-in-Process
(WIP) buffers,
as well as the Finished Board Inventory (FBI) are
represented
as queues in the model. The cards,
representing
the boards are transferred
from one
station to the next using a push system of production control after the SMT, i.e., as the batches of
boards are completed
in each station, they are
pushed to the next station without a signal from the
following station. The kanban signaling is implemented only from the FBI to the initial assembly
station (SMT). Since the push system of production
control is used after the first station, the printed
circuit boards are moved individually
from one
station to the next until they complete
all the
assembly operations.
They are batched at the last
station to transfer to the FBI in batches.

4. The simulation

results

The simulation
model developed for the line was
run one hundred times for each set of data input
over a period of one year (106250 min) and the
average was determined
for analysis. As it was

212

M. Savsar/Int.

J. Production

mentioned
above, the main goal of simulation
modeling of the assembly line was to answer the
questions raised by the management
on the minimum FB inventory
capacity and the number of
production
ordering
kanbans
needed to realize
a certain percentage
of demand met on time. In
order to answer this question, the simulation model
was run at 10 different levels of minimum number
of kanbans in the system, which was also the minimum number of batches needed in the WIP buffers
and the FBI. The number of kanbans was changed
from 50 to 105 in increments
of 5.
The results of simulation
experiments,
each
representing
the average of 100 runs, are summarized in Table 3 and graphed in Fig. 3. As it is
seen from the table and the figure, more than 99%
of demand is met on time with 90 kanbans attached
to finished and semifinished
batches of boards. In
other words, if the minimum
number of kanbans
and thus the total WIP and FBI capacity is set
equal to 90 batches, then 99.09% of demand is met
on time. In order to meet 100% of demand on time,
more than 95 kanbans are needed. Table 4 shows
the maximum
levels of WIP buffer inventories
at
all levels of minimum
number
of kanbans.
For
example, in-circuit test station had a maximum of
27 batches in its WIP at all the levels of kanbans
tested.
Figs. 4 and 5 show the fluctuations
in FBI inventory levels at two different kanban levels allowed,

Table 3
Effects of the minimum
number of kanbans
system on the percent of demand met on time

allowed

in the

Min. # of
Kanbans

Demand
met

Demand
not met

% Demand
met

50
55
60
65
70
75
80
85
90
95
100
105

1825
2080
2335
2590
2842
3082
3245
3339
3386
3412
3417
3417

1592
1337
1082
827
575
335
172
78
31
5
0
0

53.41
60.87
68.33
75.80
83.17
90.20
94.97
97.72
99.09
99.85
100
100

Economics

51 (1997) 205-214

50

55

60

65

70

75

Number of Kanbans

80

85

90

95 100105

Allowed (WIP 8~FBI)

Fig. 3. Effects of the number of kanbans allowed (total WIP and

FBI) in the system on the percentage of demand met on time.

Table 4
Maximum
Station

WIP capacities

WIP

Preassembly
In-circuit test
Final Inspection
Heat Test

realized
Maximum
2
27
19
1

Level

batches
batches
batches
batch

namely at 65 and 100 kanbans allowed. Only the

first two inventory cycles are shown completely in
the figures since the other cycles of inventory level
fluctuations
repeat in a similar behavior. These FB
inventory levels were recorded at hourly time snapshots and plotted in the figures as shown. Operation time in one day was about 7 h and the number
of working days in a week was 5. It is interesting to
see that the FB inventory
level starts to build up
until a demand arises at the end of each week (after
5 working days), at which time the inventory level is
reduced by the number of batches equal to the
demand and then a build up starts again. This
results in a typical saw tooth figure of inventory
fluctuations.
In the case of 65 minimum
kanbans,
all the
demand could not be met on time and thus, after
meeting the partial demand in a cycle, the rest of
the demand is met as the batches of boards are

M. Savsar/Int.

9 10
4
5
6
7
0
3
WORKING DAYS (5 DAYS/WEEK)

Fig. 4. Fluctuation
kanbans =65).

in final board

inventory

J. Production

11

12

level (number

of

1Do

2um
0

Fig. 5. Fluctuation
kanbans = 100).

3
4
5
6
7
8
9 10
WORKING DAYS (5 DAYS/WEEK)

in final board

inventory

11

12

13

level (number

of

assembled. Therefore, the FBI is empty for a certain

period of time until the demand is met completely,
after which time the inventory level starts to build
up. In the case of 100 minimum kanban level, all the
demand is met on time and some inventory safety
stock is left at the end of each cycle. These cases are
explained below in reference to Figs. 4 and 5.
Consider Fig. 4 which is for the case when the
number
of kanbans
were set equal to 65 units.
During the first demand cycle the FBI level reaches
to about 61 batches by the end of the first week (end

Economics

51 (1997) 205-214

213

of day 5). Thus, 61 batches of demand are met and

6 batches could not be met. The FBI level drops to
0 as the demand is met and the remaining 6 batches
are met over the next 3 h, during which time the
FBI level remains at 0. During the second demand
cycle, the FBI level reaches to about 57 batches and
thus only 57 batches of demand are met at the end
of the second week (end of day 10). The remaning
10 batches of demand are met over the following
5 h, during which time the FBI level remains at 0.
Rest of the system continues in this way.
Consider Fig. 5 which is for the case when the
number of kanbans
were set equal to 100 units.
During the first demand cycle the FBI level reaches
to about 94 batches by the end of the first week (end
of day 5). Thus, in addition
to meeting all the
demand (67 batches), 27 batches of boards remain
in the FBI. During the second demand cycle, the
FBI level reaches to about 91 batches. All of the
demand (67 batches) is met again and 24 batches of
boards remain in the FBI. The cycles that follow
continue in the same way.
The average production
cycle time, considering
equipment
failures and all random stoppages, was
found to be about 32 min per 10 boards in all cases
studied in simulation.
The average station utilizations were recorded and the results are summarized
in Table 5. It was interesting that the station utilizations were the same in all cases, i.e., the minimum
kanban
levels (also the WIP and FBI capacity)
had no effect on station utilizations
within the
limits studied. The station utilizations
included the
percent
of time that each station
was either
busy operating on a board or busy by being under

Table 5
Average station
Assembly

busy times including

station

SMT
Preassembly
Wave Solder
Insp. and Touch-Up
Post Wave
In-Circuit Test
Final Inspection
Heat Test

repair
Percent

(%)
of time busy

86
63
16
47 (each operator)
50
97 (each machine)
99
86

214

M. Savsarjlnt.

J. Production

repair. Thus, SMT utilization

of 86% means that
the SMT station was idle, i.e., waiting 14% of the
time. Thus it was either operating or under repair
86% of the time.

5. Conclusion
In this study, simulation
modeling and analysis
of a PCB assembly line is presented. A simulation
model was designed and several runs were made to
determine the number of kanbans needed to meet
the demand on time from the assembly line output.
Most of the studies on assembly line performance
are related to assembly output rate and station
utilizations without regard to the demand schedule.
While the author was studying the assembly line
described above, a different question was raised by
the management.
What is the minimum number of
kanbans
and the minimum
WIP and FBI levels
required for meeting a certain percentage
of demand on time? An assembly line can be designed
with a specified capacity to meet a specified demand. However, meeting the scheduled demand on
time needs special study and requires a certain
number of production
batches and inventory levels
to be kept particularly
if there is random fluctuation in some assembly operations.
In this study, it is shown how the simulation
modeling approach can be utilized to determine the
minimum kanbans (or batches) needed to circulate
in the system or the minimum
Finished
Board
Inventory
(FBI) as well as the WIP buffer levels
needed to meet a specified percentage of demand on
time in a real assembly line setting.
The results of this study can be further extended
by looking into the cost aspects of inventory levels
and delayed demand. In particular, one can develop a relationship
between inventory
holding cost
and demand
delay cost and then study them
through the simulation
model. This would permit
viewing how the total cost would behave if a particular cost is attached to each unit of demand not
met on time. One of the costs involved in delaying
the weekly demand in the assembly case studied
above was additional
shipment cost. For example,
when the 67 batches of demand are met on time,
a one-time shipment is made at the end of the week.

Economics

51 (1997) 205-214

On the other hand, if all the weekly demand could

not be met from the FBI at once, additional
shipments would be required during the next period
and this would result in additional costs. A balance
is needed between the inventory
holding cost and
the additional shipment costs required. This aspect
of the problem could be studied using the simulation modeling approach presented here.

References
Bard, J.F., Golany,
B., 1991. Determining
the number
of
kanbans in a multi-product,
multi-stage production
system.
Int. J. Prod. Res. 29, 881-895.
Berkley, B.J., 1993. Setting minimum performance levels for twocard kanban-controlled
lines, Int. J. Prod. Res. 31,100331021.
Chu, C., Shih, W., 1992. Simulation studies in JIT production.
Int. J. Prod. Res. 30, 257332586.
Ebrahimpour,
M., Fathi, B.M., 1984. Dynamic simulation
of
a Kanban production
inventory system. Int. J. Oper. Mgmt.
5, 5514.
Fallow, D., Browne, J., 1988. Simulating just-in-time
systems.
Int. J. Prod. Mgmt. 8(4).
Fukukawa,
T., Hong SC., 1993. The determination
of optimal
number of kanbans
in a just-in-time
production
system.
Comput. Ind. Engng. 24, 551-559.
Golhar, D.Y., Stamm, C.L., 1991. The just-in-time
philosophy:
a literature review. Int. J. Prod. Res. 29, 6577676.
Golhar,
D.Y., Sarker. B.R., 1992. Economic
manufacturing
quantity in a just-in-time
delivery system. Int. J. Prod. Res.
30, 961-972.
Huang, P.Y., Rees, L.P., Taylor, R.W., 1983. A simulation analysis of the Japanese just-in-time
technique (with kanbans) for
a multi-line, multi-stage
production
systems. Dec. Sci. 14,
3266344.
Hodgson,
T.J., Wang, D., 1991. Optimal
hybrid push/pull
control strategies for parallel multistage system: part II. Int.
J. Prod. Res. 29, 145331460.
Olhager, J., Ostlung, B., 1990. An integrated push-pull manufacturing strategy. Eur. J. Oper. Res. 45, 1355142.
Pegden, C.D., 1987. Introduction
to SIMAN. Systems Modeling
Co., Pennsylvania.
Sarker, B.R., Fitzsimmons,
J.A., 1989. The performance
of push
and pull systems: a simulation and comparative
study. Int. J.
Prod. Res. 27, 1715~1731.
Wang, H., Wang, H., 1990. Determining
the number of kanbans:
a step toward non-stock
production.
Int. J. Prod. Res. 28,
2lOll2115.
Yamashita,
H., Nakazawa,
K.. Nagasuka,
H., Suzuki, S., 1989.
Modeling and analysis of a production
line operated according to demand. JSME: Int. J. Ser. III 32, 491-497.
Zhuang, L., 1994. Towards a more economic production
and
just-in-time
delivery system. Int. J. Prod. Econom.
36,
307-313.