See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/225637216

Literature review of JIT-KANBAN system

Article  in  International Journal of Advanced Manufacturing Technology · February 2007

DOI: 10.1007/s00170-005-0340-2

CITATIONS READS

125 9,294

2 authors, including:

Ramasamy Panneerselvam
Pondicherry University
45 PUBLICATIONS   371 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Simple heuristic to minimize total tardiness in a single machine scheduling problem View project

Development of Efficient Genetic Algorithms for the Vehicle Routing Problem with Time Windows and with Time Dependency View project

All content following this page was uploaded by Ramasamy Panneerselvam on 03 January 2017.

The user has requested enhancement of the downloaded file.

Int J Adv Manuf Technol (2007) 32: 393–408
DOI 10.1007/s00170-005-0340-2
ORIGINA L ARTI CLE
C. Sendil Kumar . R. Panneerselvam
Literature review of JIT-KANBAN system
Received: 9 February 2005 / Accepted: 9 September 2005 / Published online: 22 March 2006
# Springer-Verlag London Limited 2006
Abstract In this paper, JIT (Just-In-Time)-KANBAN
literature survey was carried out and presented. The
introductory section deals with the philosophy of JIT, and
the concept involved in the push and pull system. The
blocking mechanisms in the kanban system are also
discussed elaborately. Besides these sections, the impor-
tance of measure of performance (MOP) and the applica-
tion of the same with respect to JIT-KANBAN are
presented. The recent trends in the JIT-KANBAN are
discussed under the heading “Special cases”. In this review,
100 state-of-art research papers have been surveyed. The
directions for the future works are also presented.
Keywords JIT . KANBAN . Blocking Mechanisms .
CONWIP . Measure of performances (MOP) . Simulation
1 Introduction
Just -In-Time (JIT) manufacturing system was developed
by Taiichi Ohno which is called Japanese “Toyota
production system”. JIT manufacturing system has the
primary goal of continuously reducing and ultimately
eliminating all forms of wastes (Brown et al. [5], Ohno
[54], Sugimori et al. [82]). Based on this principle,
Japanese companies are operating with very low level of
inventory and realizing exceptionally high level of quality
and productivity (Richard J. Tersine [62], James H. Greene
[30]). JIT emphasizes “zero concept” which means
achievement of the goals of zero defects, zero queues,
C. Sendil Kumar
Neyveli Lignite Corporation,
Neyveli, India
R. Panneerselvam (*)
Department of Management Studies,
School of Management, Pondicherry University,
Pondicherry 605 014, India
e-mail: panneer_dms@yahoo.co.in
zero inventories, zero breakdown and so on. It ensures the
supply of right parts in right quantity in the right place and
at the right time. Hence, the old system of material
acquisition and, buyer and seller relationships are changed
to new revolutionary concepts (Womack et al. [91],
Womack and Jones [92], Markey et al. [45]). Similarly,
JIT becomes an inevitable system at plant level, which
integrates the cellular manufacturing, flexible manufactur-
ing, computer integrated manufacturing and Robotics
(Schonberger [63], Golhar [12]).
Due to the technological advancement, the conventional
method of push production system linked with Material
Requirement Planning (MRP) was changed to pull type JIT
production system to meet out the global competition,
where the work-in-process (WIP) can be managed and
controlled more accurately than the push- production
system (Mason Paul [46]).
KANBAN system is a new philosophy, which plays a
significant role in the JIT production system. Kanban is
basically a plastic card containing all the information
required for production/assembly of a product at each stage
and details of its path of completion. The kanban system is
a multistage production scheduling and inventory control
system. These cards are used to control production flow
and inventory. This system facilitates high production
volume and high capacity utilization with reduced produc-
tion time and work-in-process.
The objectives of this paper are as listed below
1) Critical review of JIT literature.
2) Segregating the different research articles of JIT.
3) Exploring the recent trends in JIT-Kanban system and
deriving directions for future research.
In this paper, the articles are reviewed and an appropriate
classification is presented.The kanban study was made
elaborately, since it acts as a basic communicator and feed-
back agent to the JIT system. Push and pull system,
principle of operation of kanban cards, Blocking mecha-
nism, Toyota’s formula, and the measures of performances
(MOP) are also discussed in this paper. The latest trends in
JIT-Kanban system are also addressed separately under the
394
heading “Special cases”. Finally, the directions for future
researches are presented.
2 Push and pull systems
Push and Pull system are two types of production systems,
which operate equally in opposite sense and have their own
merits and demerits (Monden [50], Villeda Ramiro et al.
[89]).
Push system It is a conventional system of production.
When a job completes its process in a workstation, then it
is pushed to the next workstation where it requires further
processing or storing. In this system, the job has a job card
and the job card is transferred stage by stage according to
its sequence. In this method, due to unpredictable changes
in demand or production hinder-ness, the job happens to
deviate from its schedule and it causes accumulation of
work-in-process inventory. Hence, inventory planners
pessimistically fix the safety stock level on the higher
side. A schematic representation of the push system is
shown in Fig. 1. In Fig. 1, WSj is the jth workstation and
the product line consists of n workstations.
Pull system A pull type production system consists of a
sequence of workstations involving value addition in
each workstation (WS). In the pull system, from the
current workstation (j), each job is withdrawn by its
succeeding workstation (j+1). In other words, the job is
pulled by the successive workstation instead of being
pushed by its preceding workstation. The flow of parts
throughout the product line is controlled by Kanban Cards
(Turbo [87]). In practice, these kanban cards can be either
“single-card system” or “two-card system”. Each work-
station has an inbound stocking point and an outbound
stocking point. The primary advantage of the pull system
is the reduced inventory and hence the associated cost of
inventory reduction. A schematic view of the pull system
with two workstations and store is shown in Fig. 2.
A kanban system operates only with single card is
called production order kanban (POK) (J. Berkley [4],
Sarathapreeyadarishini et al. [78]). If the distance between
the consecutive workstations is very short, a single buffer
mode is made available between the workstations. This
buffer mode acts as both outbound buffer for the current
workstation j and inbound buffer for the succeeding
workstation j+1, respectively. A schematic diagram of a
single-card system is shown in Fig. 3. In the two-card
system, where the distance between the two consecutive
work stations are more, each work station will have
separate inbound buffer and outbound buffer (Kimura O. et
al. [36], Hemamalini et al. [21]) and the cards are called as
Production Order Kanban (POK) and Withdrawal Kanban
WS1 WS2 ooo WSj ooo WSn STORE
Fig. 1 Push system
Request for items Request for items
WS 1 WS 2 STORE
Items movement Items movement
Fig. 2 Pull system
(WK), respectively. A schematic diagram of a two-card
system is shown in Fig. 4.
2.1 Operation of two-card kanban system
The two-card kanban pull system which works in
the Assembly/Manufacturing line is elaborated by
Panneerselvam [56], Edward J. Hay [17], Kimura and
Terada [36], Hunglin Wang et al. [25] and Hemamalini et
al. [21] and Shahabudeen et al. [76]. Basically it has plastic
cards, which give information about the parts and also
things to be done. The production order kanban (POK) is a
production order, which instructs the preceding work-
station to produce the required number of units. The
withdrawal kanban (WK) gives the message to the
succeeding process about the number of units it should
withdraw.
The simple steps adopted in kanban system are as
follows
1) The container of the succeeding workstation j+1 is
moved to the preceding workstation j with the
withdrawal kanban (WK) and placed it in its output
buffer.
2)
a) Consequently it pulls the parts from output buffer of
the workstation j and detach the production order
kanban (POK) attached to those parts and then places
the POK in the POK-post of the workstation j.
b) Work station j starts its production as per the
production order in its POK post.
3) The container along with the parts and WK moves
again to its succeeding workstation j+1. Then it
delivers the parts to the input buffer of the workstation
j+1 and places the WK to the WK-post of the
workstation j+1.
POK Only card movement
POST
Card + Parts movement
WSj WS j+1
BUFEER
Fig. 3 Schematic diagram of a single card system
 395
POK WK 3.1.1 Blocking due to part type
POST POST
(j) (j+1)
This type of blocking occurs due to restriction in the
number of parts (containers) that can be stored in the buffer
POK WK
between workstation j and the workstation j+1. The
workstation j will not process the particular part p, since
there is no reserved space in the buffer storage for the
WS WK+ Parts WS particular part type.
j j+1 Let Q(p, j, j+1) be the maximum number of units
Output Buffer Input Buffer (container) of part type p that can be stored in the buffer
of of storage between the workstation j and the workstation j+1.
Workstation j Workstation j+1 Then the workstation j can process the p type parts, only if the
actual number of units (container) of the part type p in the
WK buffer storage less than Q(p, j, j+1); otherwise, the work-
station j is blocked due to part type p alone. The workstation
Fig. 4 Schematic diagram of two card system can process any other part type provided that workstation is
not blocked with respect to that part type.
4) When the parts in the containers of the workstation j+1
are fully used, then the steps from 1 to 3 are repeated.
3.1.2 Blocking due to queue size

3 Blocking mechanisms This type of blocking occurs due to restriction in the total
number of containers of all part types in the buffer between
Each workstation of a production/assembly line requires workstation j and the workstation j+1. The workstation j
sufficient space for storing parts in its output buffers. When will not process any part type if there is no space in the
the buffer capacity of a workstation is fully occupied, no buffer storage between the workstation j and the work-
further storage is possible. Because of this fact, the station j+1, irrespective of part type and container.
workstation can not release the parts and hence, it can Let Q (j, j+1) be the maximum number of containers
not process components. This condition is called “Block- irrespective of the part types that can be stored in the buffer
ing”. The blockings are categorized according to the types storage between the workstation j and the workstation j+1.
as presented in Table 1. Then the workstation j can process part types, only if the
actual total number of containers in the storage between the
workstation j and the workstation j+1 is less than Q(j, j+1);
3.1 Single card -instantaneous Otherwise, the work station j is said to be blocked due to
the queue size constraint.
As discussed earlier, if the workstations are situated closer
to each other, the output buffer of the workstation j and the
input buffer of the workstation j+1 are one and the same. 3.1.3 Dual blocking mechanism
Under such situation, a single card instantaneous kanban is
used. Berkley [4] and Sharadhapreeyadarishini et al. [77] If both the above blocking mechanisms operate simulta-
have discussed the blocking mechanism of single card type neously, then it is called Dual blocking mechanism.
in detail. The work station j is said to be blocked if the actual
number of units (containers) of the part p in the buffer
storage between the workstation j and the workstation j+1
is equal to Q(p, j, j+1) and the actual total number of
containers in the buffer storage between the workstation j
and the workstation j+1 is equal to Q(j, j+1).
Table 1 Categories of blocking mechanisms Subsequently, when a container of the part type p is
taken by workstation j+1, then the blocking is released and
Single Card -Instantaneous Two Card - Non Instantaneous
the workstation j can start processing the part p. If the work
1) Blocking due to part-type. 1) Blocking due to part-type. station j+1 takes the container of any part other than that of
2) Blocking due to queue size. 2) Blocking due to queue size. p, then the work station j is still blocked with respect to part
3) Dual blocking mechanism. 3) Dual blocking mechanism. p and it is not blocked with respect to other part types.
Blocking mechanism Operative
on Material Handling.
4) Blocking due to part-type.
3.2 Two card- non-instantaneous
5) Blocking due to queue size.
6) Dual blocking mechanism.
If the distance between consecutive workstations is more,
there will be independent input and output buffer points for
396
each workstation. In this system, the blocking can occur
due to stagnation of parts in the output buffer of that
workstation. Berkley J. [4] et al. [21] have studied this type
of blocking.
3.2.1 Blocking due to part type
This type of blocking occurs due to restriction in the
number of parts (containers) that can be stored in the output
buffer of the workstation j. The workstation j can not
process the particular part p, since there is no reserved
space for the part type p in the output buffer of the
workstation j.
Let Q(p, j) be the maximum number of units (containers)
of the part type p that can be stored in the output buffer
storage of workstation j. Then the workstation j can process
the parts of the part type p, only if the actual number of
units (containers) of p in the output buffer storage of the
workstation j is less than Q(p, j); otherwise, the workstation
j is blocked due to the part type p alone. The workstation j
can process parts of any other part type provided that the
workstation is not blocked with respect to that part type.
3.2.2 Blocking due to queue size
This type of blocking occurs due to restriction in the total
number of containers of all part types in the output buffer of
the workstation j. The workstation j will not process any of
the part types since there is no space in the output buffer
storage of the workstation j, irrespective of part type and
container.
Let Q(j) denotes the maximum number of containers
irrespective of part type that can be stored in the output
buffer storage of the workstation j.
Then the workstation j can process parts only if the
actual total number of containers in the output buffer of the
workstation j is less than Q(j); otherwise, the workstation j
is said to be blocked due to the queue size constraint.
3.2.3 Dual blocking mechanism
If both of the above blocking mechanisms operate simul-
taneously, then it is called dual blocking mechanism.
The workstation j is said to be blocked if the actual
number of units (containers) of part type p in the output
buffer of workstation j is equal to Q(p, j) and the actual total
number of containers in the output buffer of workstation j is
equal to Q(j).
Subsequently, when a container of part type p is taken to
the input buffer of the workstation j+1, the blocking will be
released and the workstation j can start processing the part
type p. If the input buffer of workstation j+1 takes a
container of parts other than the part p, then the workstation
j is still blocked with respect to the part type p.
3.3 Blocking mechanisms operative on material
handling
Material handling operation between the workstation j and
the workstation j+1 can be blocked due to part type, queue
size or both. This is similar to the above types but the
blocking is due to Material Handling (MH) between output
buffer of the workstation j and the input buffer of the
workstation j+1. This was studied by Berkley J. [4] and
Hemamalini et al. [21].
3.3.1 Blocking mechanism due to part type
This type of blocking occurs due to restriction in the
number of parts (containers) that can be stored in the input
buffer of the workstation j+1.
Let M(p, j+1) denotes the maximum number of units
(containers) of part type p that can be stored in the input
buffer storage of the workstation j+1.
Then, materials handling is permitted from the output
buffer of work station j to the input buffer of workstation
j+1, if the actual number of units (containers) of part p in
the input buffer of work station j+1 is less than M(p, j+1).
3.3.2 Blocking mechanism due to queue size
This type of blocking occurs due to restriction in the total
number of containers of all part types that can be stored in
the input buffer of the workstation j+1.
Let M(j+1) be the maximum number of containers
irrespective of part types that can be stored in the input
buffer storage of workstation j+1. Then, the material
handling is permitted from output buffer of the workstation
j to the input buffer of workstation j+1 only, if the actual
total number of containers in the input buffer of work
station j+1 is less than M(j+1).
3.3.3 Dual blocking mechanism
If both blocking mechanisms discussed in sections 3.3.1
and 3.3.2 operate simultaneously then it is called dual
blocking mechanism. The material handling operation is
said to be blocked, if the actual number of units (contain-
ers) of part type p in the input buffer of the workstation j+1
is equal to M(p, j+1) and the actual total number of
containers in the input buffer of the workstation j+1 is
equal to M(j+1). Subsequently, when a container of the part
type p is taken from the input buffer of workstation j+1, the
blocking will be released and the material handling starts to
clear the parts from the output buffer of the work station j.
Now, the workstation j can start processing the part p. If the
workstation j+1 takes a container of parts other than that of
part p from the input buffer of the workstation j+1, then the
material handling is not possible for the part p. So the
workstation j is continued to be in blocked state with
 397

respect to the part p. However, the workstation j is not Table 2 Factors used by various researchers
blocked with respect to other part types. Sl. Factors used The reference numbers
No for MOP of research articles which
use the MOP
4 Toyota’s kanban formula
1 Average WIP [2, 6, 15, 48, 76, 79, 89, 90, 98]
The formula used by Toyota Motor Company to determine 2 Demand [6, 15, 76]
the number of kanbans is called Toyota formula. (Berkley 3 Fill Rate [6]
[4], Chan [6], Henry et al. [10], Hunglin Wang et al. [25], 4 Average kanban [74, 77, 98]
Ohno et al. [53], Monden Y. [50], Philipoom et al. [60] and waiting/queue time
Yavuz et al. [98]). The Toyota’s kanban formula is 5 Average Flow [2, 6, 22, 31, 51, 61, 76, 77, 98]
presented below. (production lead) time
6 Average setup/process [98]
D Lð1 þ αÞ
K
time ratio
C 7 Average input/output [79, 89]
inventory
where, 8 Mean cumulative [48, 74, 88, 90, 98]
K is the number of kanbans, throughput rate
D is the demand per unit time, 9 Mean line utilization [48, 89, 98]
L is the lead-time, 10 Mean demand [98]
α is the safety factor and satisfaction lead time
C is the container capacity 11 Mean staging delay of job [49, 51]
From these literatures, it was noted that the lead-time 12 Mean Tardiness [2, 22, 51, 61, 79]
includes waiting time, processing time, conveyance time 13 Weighted earliness of the [22, 61]
and kanban collecting time. The safety stock serves as a job
buffer against variations in both supply and demand. Henry
et al. [10] has suggested some practical values for the
variables C and α. The value of C is limited to a maximum frequently used as performance measures. Some important
of 10% of demand and α is a policy variable, which is definitions for the factors, which are used in different
decided by the management up to 10% of the demand. The MOPs by various researchers, are discussed below.
variable K is the number of kanbans, which is related to the Yavuz and Satir [98] have used seven factors in their
stock. If the value of K increases, the stock of the parts also study, which are as presented below.
increases. As a result, idle stock occurs. Similarly, if the
1) Mean Cumulative Throughput Rate: It is the ratio of
value of K decreases, the stock of parts also decreases and
total satisfied demand to the total generated demand.
shortage occurs. Hence, the JIT production system applies
2) Mean Total Production Lead Time: It is the amount of
trade-off between the above parameters to find the optimum
time spent by a job from entering the system to until
number of kanbans. Many researches have been carried out
completion of all operations, averaged over all
to find the optimum number of kanbans using different
completed job.
methodologies and tools such as simulation, queuing
3) Mean Total Demand Satisfaction Lead Time: It is the
models, mathematical models, Artificial Intelligent ap-
time interval between arrival of the demand and
proach and so on. From the Toyota’s empirical equation,
satisfaction of the demand.
one can find the number of kanbans required for the system.
4) Mean Utilization of Line: It is the mean utilization of
the last station in the line.
5) Mean Setup/Run Time Ratio of Line: It is the ratio
5 Measure of performance (MOP)
between the setup time and the run time of last station.
6) Mean Total WIP Length: It is the mean of all in-
For any system, the efficiency is measured through a
process-inventory levels for the products excluding
function of related parameters/ factors. Hence these factors
finished goods (FG).
must obviously establish close relationship with the
7) Mean Total Waiting Time: It is the waiting time of all
focused problem. These factors individually or jointly repre-
products in all processes and finished goods inventory
sent a performance. Blair Berkly J. [4] has given a note on
(FGI).
workstation performance in kanban controlled shops in
terms of average inventories, quality and the ability to meet A general purpose analytical model to evaluate the
the demands. Our study reveals that various researchers performance of multistage kanban controlled production
have used thirteen factors and they are shown in Table 2. system was developed by Di Mascolo et al. [15]. The
From Table 2, it is inferred that the average work-in- performance measures used by them are percentage of
process (WIP), average flow time, mean cumulative demand for back-order, average waiting time of back-
throughput rate and weighted earliness of the job are order and average work-in-process.
398

A simulation experiment to evaluate the relative effective- Table 3 Results of the study by Chan [6]
ness of various rescheduling policies in capacity-constrained, MOP Pull Hybrid Hybrid
JIT make-to-stock production environment is examined by (Single (Single (Multi-product)
Kern et al. [34]. Three performance measures analyzed by product) product)
them are average finished goods inventory, total units of sales
lost, and measure of schedule instability. Jing-Wen Li [31] 1) Fill rate Decrease Decrease Increase
has measured three factors for shop performance which are 2) In-process- Increase Increase Increase
average work-in-process (WIP) inventory, average flow time inventory
and average set up time to processing time ratio (ASOTR), 3) Manufacturing Increase Increase Decrease
which is the ratio of total amount of time spent for setting up lead time
machines to the total amount of time spent for processing
parts averaged over all machines. Uday S. Karmarker [88]
used throughput rate for total work performance. In another
study, the priority rule assignment was checked by the 6 Literature review
following factors by Nabil R. Adam et al. [51].
Golhar et al. [12] have classified the JIT literature as
1) The lead time of a job
elimination of waste, employee participation, supplier
2) The flow time of the job
participation and total quality control.
3) The staging delay of a job
A similar work was done by Berkly [4] for kanban
4) Mean Tardiness
production process. He has selected 24 elements in the
Hemamalini et al. [22] considered the objective function kanban production system as operational design factors.
to minimize the sum of weighted flow time, weighted In this section, the different topics associated with “JIT-
earliness of jobs and weighted tardiness of containers. KANBAN” studied by various researchers have been
Shahabudeen et al. [76] used an universal test which may grouped and presented as shown in Fig. 5. The Table 4
be suited for the MOP in any JIT system, which are shows the reference numbers of the articles with respect to
percentage zero demand (PZD), mean lead time (MLT) and the classifications shown in Fig. 5.
mean total WIP (MTW) as explained below. Obviously, most of the researchers were focusing on the
determination of number of kanbans and determining
1) Percentage zero demand: It is the percentage of total
corresponding solutions by using suitable models and
demand immediately satisfied to the total generated
tools. Some authors have developed simulations model and
demand.
meta-heuristics like, genetic algorithm (GA), tabu search
2) Mean lead time: It is the sum of the waiting time,
(TS), and simulated annealing (SA) for JIT-Kanban for
processing time and moving time averaged per station.
better solutions.
It is also called as mean flow time.
The Table 5 shows the number of articles dealt in
3) Mean total WIP: It is the average number of kanbans
different periods. From, Table 5, it is clear that, during last
waiting for each part type at each workstation.
two 5 years period (1996-2000 & 2001-2005), the number
Here, PZD is a maximization measure and, MLT and of researches are more. Further, more researches have been
MTW are the minimization measures and hence the sum of done in empirical theory, flow shop, simulation, variability
the objective MOPs is changed as Zmax (a1PZD +a2 RMLT and its effects, CONWIP and special cases. Many
+a3 RMTW), where a1, a2 and a3 are weights of the researchers have worked in JIT system with different
respective measures and, RMLT and RMTW are modified objectives. Here, the authors have grouped some important
values of MLT and MTW, respectively.
Chan F.T.S. [6] has done a work on how the MOP JIT
changes in different production systems, while increasing
the kanban size. The measures of performance taken by SPECIAL CASES
him are as listed below
1) Unsatisfied order, which is the difference between the KANBAN CONWIP
actual number of unit produced and the level of SCM
demand.
2) Manufacturing lead-time, which is the time between POLCA
the customer order and the completion of order. VARIABILITY &
3) In-process-inventory is the total number of work-in- ITS EFFECTS
process (WIP) inventory in units excluding finished FLOW SHOP
DIFFERENT MODELS
goods inventory. ASSEMBLY MATHEMATICAL
4) Fill rate is the percentage of demand satisfied. LINE QUEUING
MARKOVIANS
SIMULATIONS
The results of his study as a function of the kanban size BATCH COST MINIMIZATION
are shown in Table 3.
Fig. 5 Flowchart showing the classification of literature review
 399
Table 4 Details of classification of review articles Table 6 Objective based classification and their references
Area of Research Reference numbers of related Articles Classification Reference numbers of articles

JIT [4, 12] A. Principles of [21, 50, 55, 59, 96]

Kanban-Empirical theory [10, 33, 50, 59, 71, 93, 95] JIT-Kanban system
Flow shop [7, 22, 58, 61, 77, 78] B. Operating Factors [4, 12, 80]
Assembly line [16, 89, 94] C. Design of Kanban [1, 10, 13, 19, 26, 52, 53, 66, 71, 74, 75,
Batch Production [35, 86] System 93]
System D. Performance [7, 24, 33, 36, 48, 58, 73, 81, 89, 90, 98,
Modeling Approach: [3, 36] behaviour 99]
Mathematical E. Sequencing [18, 22, 57, 58, 61, 77, 78, 94]
Queuing [73, 99] & Scheduling
Marko-chain [14, 28, 52, 90] F. Inventory/Buffer [3, 68, 86]
Simulation [1, 9, 13, 19, 26, 66, 67, 74, 75] Control
Cost minimization [53, 68, 72]
Variability and its effects [7, 24, 48, 89, 98]
CONWIP [8, 44, 55, 69, 70, 79, 96]
POLCA [69] Karmarker and Kekre [33] have concluded from their
SCM [18, 29, 47, 80]
studies that the reduction in container size and increase in
Special Cases [6, 38, 40, 41, 43, 60, 69, 84, 85, 97, 100]
number of kanbans lead to better results. Many researchers
were interested in finding the optimal number of kanbans.
The Toyota formula is very much useful in determining the
optimal number of kanbans.
objectives of the researches into six headings as shown in Co Henry et al. [10] used the Toyota formula and also
Table 6. From Table 6, it is clear that the following investigated the safety stock allocations in an uncertain
objectives attracted more researchers. dynamic environment. A similar work was considered by
Sarkar et al. [71] to find number of kanbans between two
– Design of kanban system
adjacent workstations. Yale T. Herer et al. [95] presented a
– Performance behaviour
study for kanban system, CONWIP and buffered produc-
– Sequencing and scheduling
tion lines. In this study, they incorporated a non-integral
approach using simulation. The use of non-integral
approach helps production planners to obtain discrete
6.1 Empirical theory number of kanbans.
Woolsey et al. [93] have developed a simple spreadsheet
In the paper by Monden Y. [50], a comprehensive optimization program to determine the corresponding
presentation of Toyota production system is given. A number of kanbans with respect to user-defined safety
successful kanban system will drastically reduce the stock levels and other values. It gives a close-form of
throughput time and lead time (Philipoom et al. [59]).

Table 5 Details of researches in different periods

Area of Research 1980-1985 1986-1990 1991-1995 1996-2000 2001-2005 Total

JIT 2 2
Kanban- Empirical theory 1 2 3 6
Flow shop 1 4 1 6
Assembly line 1 1 1 3
Batch 1 1 2
Modelling Approach: Mathematical 1 1 2
Queueing 2 2
Markovians 1 2 1 4
Simulation 1 1 5 2 9
Cost minimization 1 1 1 3
Variability and its effects 1 1 3 5
CONWIP 2 3 2 7
POLCA 1 1
SCM 4 4
Special cases 1 1 2 7 11
Total 3 10 11 23 20 67
400
solution to the problem. This means that an answer for any
problem size may be instantaneously obtained.
6.1.1 Flow shop
Kanban system is widely implemented in repetitive
manufacturing environment. For a single card operational
system, Sharadhapriyadarishini et al. [77] have developed
two heuristics and proved that these are more efficient.
Saradhapriyadarishini et al. [78] have proposed a recursive
equation for scheduling the single card kanban system with
dual blocking. They proposed a heuristic with twin
objectives of minimizing the sum of total weighted time
of containers and weighted flow time of part-types.
Rajendran [61] has done a work on two card flow shop
scheduling with n part-types. In this paper, mathematical
models for time tabling of containers for different problems
have been formulated. Then, a heuristic was developed to
minimize the sum of weighted flow time, weighted earliness,
and weighted tardiness of containers. Hemamalini et al. [22]
have done similar work. In this work, the heuristic developed
is simulated annealing algorithm. This is compared with
random search method. In these papers, the comparisons are
done only based on mean relative percentage increase.
Instead of this approach, comparisons based on complete
ANOVA experiments would provide reliable inference.
Peter Brucker et al. [58] have carried out research on
flow shop problem with a buffer of limited capacity
between two adjacent machines. After finishing the
processing of a job on a machine, either the job is to be
processed on the following machine or it is to be stored in
the buffer between these machines. If the buffer is
completely occupied, the job has to wait on its current
machine but blocks this machine for other jobs. In this
paper, they determined a feasible schedule to minimize the
makespan using tabu search. The results of the problem
using tabu search were compared with that of benchmark
instances. The comparisons are done only based on relative
improvements. Instead of this approach, comparisons
based on complete ANOVA experiments would provide
reliable inference.
6.1.2 Assembly line
Assembly lines are similar to the flow shops in which
assembly of parts are carried out in a line sequence. In a
multi product assembly line, the sequencing of the jobs is a
challenging task. Drexl et al. [16] considered an assembly
line sequencing mixed model problem. It is a combinatorial
problem. They formulated this combinational problem as
integer programming model. This model can be used only
for small size problems due to the limitations of operations
research software with respect to handling the number of
variables and constraints, which are present in the integer-
programming model. Xiaobo et al. [94] have considered
similar work on mixed model assembly line sequencing
problem with conveyor stoppages. They proposed branch
and bound algorithm, and simulated annealing algorithm
for finding the optimal solution and sub-optimal solution of
the mixed-model sequencing problem, respectively to
minimize the total conveyor stoppage time. The branch-
and-bound method was devoted to find the optimal
solution of small-sized problems, whereas the simulated
annealing method was used to cope with large-scale
problems to obtain a good sub-optimal solution. Future,
research on simulated annealing applied to this problem
can be directed to establish a better seed generation
algorithm. However, the practitioner should spend con-
siderable time in fixing the parameter called temperature
(T) in the simulated annealing algorithm by trail and error
method before actually solving the problem.
6.1.3 Batch production system
In a batch production system, the switching over from one
product to other product depends on many factors such as
stock reaching to the threshold level, different priority
schemes, economical setups, etc. Tafur Altiok et al. [86]
have dealt this issue differently for the pull type
manufacturing system with multi product types. In this
paper, they developed an iterative procedure to approxi-
mately compute the average inventory level of each
product as finished goods using different priority schemes.
In this paper, the demand arrival process is assumed to be
a poisson distribution and processing times and the set-up
times are arbitrarily distributed. But, in practice, the
processing times may follow other distributions, viz.,
normal, uniform, exponential, etc. which are not experi-
mented in this paper. Khan et al. [35] addressed the
problem of manufacturing system that procures raw
materials from vendors in lot and convert them into
finished products. They estimated production batch sizes
for JIT delivery system and designed a JIT raw material
supply system. A simple algorithm was developed to
compute the batch sizes for both manufacturing and raw
material purchasing policies.
6.2 Modeling approach
Modelling approach aims to obtain the optimal solution.
This subsection reviews different modeling approaches.
6.2.1 Mathematical model
Kimera and Terada [36] have developed a mathematical
model in the area of kanban system. They have given a
basic balance equation for multi stage systems, which
shows how the fluctuation of final demand influences the
fluctuation of production and inventory volumes. Bitran
and Chang [3] have designed an optimization model for the
kanban system. The model is intended for a deterministic
multi-stage capacitated assembly-type production setting.
In this paper, a non-linear model developed by them is

converted into a linear model with deterministic demand.
This deterministic model is designed to find the choice of
the number of kanbans to be used at each stage of a given
problem and to control the level of inventory. But this
analysis does not include uncertainties directly. Hence, the
utility of this model is very much limited.
6.2.2 Queuing model
Seki et al. [73] have designed a single-stage kanban system
with poisson demand arrivals. The system is formulated as
a queuing system under piecewise constant load, and a
numerical method by transient solutions of the queue is
applied. This method, which shows the transient behavior
of the kanban system, gives a better result. Yoichi Seki et
al. [99] did similar work on the single stage kanban system
with poisson demand and erlang production times. The
objective of this work is to determine the number of
kanbans, when a change of load to the system is planned.
They mainly proposed a numerical method by transient
solutions of the queueing system which was developed
under piecewise constant load. This method also shows
that the transient behavior of the kanban system operates
better with other parameters. In this paper, the load
distribution is assumed to be piecewise linear. Instead, it
can be assumed as a continuous distribution and the
corresponding results using simulation can be compared
with `the results of this paper.
6.2.3 Markovians model
Vito Albino et al. [90] studied a model of kanban controlled
manufacturing system based on Markovian assumption.
An approximate approach was developed to solve the
model, which permits reliable evaluation of performance in
terms of throughput time and work-in-process (WIP).
Further, they validated the results using discrete-event
simulation applied to their problem. It was observed that
the results of the approximation approach did not deviate
much from that of the simulation approach. The errors were
always within 5% even for moderate size problems with 20
stages. The comparisons made in this paper were based on
absolute value of percentage relative errors. Instead of this
approach, they should have done comparisons through a
carefully designed ANOVA experiments. Nori and Sarkar
[52] have modeled the kanban system using Markov-chain
to determine the optimum number of kanbans between
adjacent workstations.
Deleersnyder et al. [14] have modeled a blocking
situation in the queues of the kanban system using discrete
time Markovian chain to study the effect of number of
kanbans, machine reliability, processing time and demand
variability. Markham et al. [28] formed a procedure based
rule induction approach for determining the number of
kanbans and other factors in JIT. They applied classifica-
tion and regression tree (CART) technique to generate the
production rule, based on decision trees. This system
401
approach involves 3 steps methodology, viz., 1) data
collection, 2) formation of decision tree, and 3) interpre-
tation of decision tree. This method helps to set kanban
levels under high demand variability. The results show that
rule induction using CART is a viable solution to the
knowledge acquisition bottleneck. Hence, an extended
work on knowledge acquisition for this domain will be a
significant contribution to literature.
6.2.4 Simulation based studies
There are many simulation softwares available in the
market, such as GPSS, Q-GERT, SLAM-II, SIMAN,
SIMSCRIPT, EXTEND, ARENA, and SIMULINK. Sim-
ulation uses the attributes/parameters of a problem to arrive
the results. As for as designing of kanban system, a basic
simulation study was done by Davis et al. [13] and Gabriel
et al. [19] to determine the number of kanbans. In another
work by Rudi De Smet et al. [67], a simulation model was
developed to study the feasibility of plans to produce some
subparts of the product in a kanban-controlled manner to
determine the operational parameters such as number of
kanbans and container size. This feasibility study was
carried out for two situations, namely (1) all subpart types
are produced in a kanban controlled manner and (2) only
the production of fast-movers on two (out of three)
machines is kanban controlled. The result assures that the
kanban control is the best method for fast moving parts.
In a kanban control system, the main decision
parameters are the number of kanbans and lot size. Alabas
et al. [1] developed three-meta heuristics viz., genetic
algorithm (GA), simulated annealing (SA) and tabu search
(TS) coupled with a simulation model to find the optimum
number of kanbans with the minimum cost. In addition, a
neural network metamodel was developed and compared
with the heuristic procedures in terms of solution accuracy.
They found that the tabu search requires less computational
efforts when compared to the other two meta-heuristics and
the neural network meta-model. In a similar work by
Hurrion R.D. [26], simulation and neural network meta-
model have been used for designing the kanban system. In
this paper, an approximate solution is found using neural
network meta-mdoel and then it is used as the starting point
in simulation to find the optimum number of kanbans of a
manufacturing system. Actually, the word “optimum”
should have been avoided in his paper, because neither
the proposed meta-model nor the simulation approach will
give optimal number of kanbans. The optimum number of
kanbans may be called as the minimum number of kanbans.
In this context, an attempt has been made by
Shahabudeen et al. [75] to set the number of kanbans as
well as lot size at each station using simulated annealing
algorithm. A simulation model with a single-card system
has been designed and used in the analysis. A bi-criterion
objective function comprising of mean throughput rate and
aggregate average kanban queue, has been used for
evaluation. In another work of them (Shahabudeen and
Krishnaiah [74]), they have set the number of production
402
kanbans and withdrawl kanbans at each workstation, and
lot size using genetic algorithm (GA). The solution of the
genetic algorithm is found to be better than the random
search procedure. They concluded that the genetic algo-
rithm gives better solution for the assumed kanban system.
A paper by Royce O. Bowden et al. [66] describes the
use of evolutionary programming (EP) integrated with a
simulation model of manufacturing system to determine
the minimum number of kanbans and corresponding
production trigger values required to meet the demand. In
this paper, the inference is drawn for each measure, based
on single replication under each solution-technique. The
authors could have designed a single factor ANOVA
experiment for each measure in which “Solution Tech-
nique” as the factor, with desirable number of replications
to obtain reliable inference of their simulation study.
Christos G. Panayioton et al.[9] have developed a simu-
lation based algorithm for determining the minimum
number of kanbans in a serial production system in order
to maximize the throughput rate and minimize work-in-
process inventory. The finite perturbation analysis (FPA)
technique was used in the simulation and to get sensitivity
results. They have considered single product in the
production line. But, in most of the cases, production
lines will be manufacturing multi-products. The assump-
tions of arbitrary arrival and service process distributions
limit the scope of application of this paper in practice.
6.2.5 Cost minimization model
Ohno et al. [53] proposed an algorithm to determine the
optimal number of kanbans for each of the two kinds of
kanban (production ordering and supplier kanbans) under
stochastic demand. An algorithm was devised for deter-
mining the optimal number of kanbans that minimizes the
expected average cost per period. Since, no safety stock is
assumed in this paper, this can be regarded as a procedure
for determining the safety stock also. Sarkar et al.[72]
studied a multi stage kanban system for short life-cycle
product in the market. In this research, the problem is to
find optimally the number of orders for raw-materials,
kanbans circulated between workstations, finished goods
shipments to the buyers, and the batch size for each
shipment (lot) with minimum total cost of the inventory. A
cost function was developed based on the costs incurred for
the raw materials, the work-in-process and the finished
goods. The optimal number of raw material orders that
minimizes the total cost is obtained first, which is then used
to find the minimum number of kanbans, finished goods
shipments, and the batch sizes of shipments. This paper
discusses a stage-wise optimization. Instead, a fully
integrated approach may be followed. Further, this paper
considers single product, with constant production rate at
each workstation in a serial production line. So, the work
may be extended for multi-product with varying production
rate at each workstation in an assembly-type production.
During preventive maintenance, a JIT buffer is needed
so that the normal operation will not be interrupted. The
optimal JIT buffer level is determined from a cost analysis
using trade-off between the holding cost per unit of time
and the shortage cost per unit time such that their sum is
minimized (Salmark et al. [68]).
6.3 Variability and its effects
Mehmet Savsar et al. [48] studied a simulation model to
investigate the effect of different operational conditions,
including kanban withdrawal policies on three perfor-
mance measures of JIT, viz., average throughput rate,
average station utilization and average work-in-process.
Unlike other simulation studies that use exponential or
truncated normal distribution, this model uses Erlang and
Gama distribution. It is observed that the throughput rate as
well as the average station utilization is significantly
affected by the variability in processing time and demand
intervals. They proposed two types of kanban withdrawal
cycles, namely fixed withdrawal policy and variable
withdrawal policy. Under the fixed withdrawal policy, the
time interval between consecutive visits of a part-carrier to
a workstation for kanban removal is fixed, but the order
quantity (number of kanbans carried) is variable whereas
under the variable withdrawal policy, the time interval
between consecutive visits of a part-carrier to a workstation
for kanban removal is variable, but the order quantity is
fixed. As an extension of this work, the effects of different
combinations of the two kanban withdrawal policies and
number of kanbans between workstations, on the perfor-
mance measures can be compared.
Huang et al. [24] have found that overtime required will
be increased when the variation in processing time is
increased. Also, they emphasized that a kanban system
would not be effective with high variable processing or set
up time. Villeda et al. [89] performed a simulation study for
a final assembly consisting of “3 sub-assembly lines and 4
stages” repetitive production systems with kanbans. They
concluded that improved productivity obtained through
unbalancing the processing time at all workstations
increases directly with the variability in the final assembly.
Chaturvedi and Golhar [7] simulated a kanban based flow
production line for a product in nine sequentially arranged
workstations. They observed that the system performance
was worst for exponential processing time distribution and
variability affected station utilization, throughput time and
WIP inventory. Yavuz and Satir [98] have studied the
simulation of multi-item, multi-stage flow line operating
under the JIT philosophy with a two-card kanban tech-
nique. The flow line produces four products through five
stations. This study uses partial factorial design for
experimentation. Seven experimental clusters are designed,
each composed of at most three factors. The F ratios and
the degrees of freedom of the model are obtained from
multi-variate analysis of variance (MANOVA). They found
that decrease in lot size reduces mean length and waiting
times in work-in-process points at all kanbans levels. An
increase in the uncertainty of demand arrival rates and
demand sizes increases the probability of sudden over-

loading. An increase in the coefficient of variation in
processing times brings about higher line utilization and a
decrease in throughput rate. The scheduling rules tested
in this paper are found to yield no significant differences in
the utilization of line and on the behaviours of work-in-
process. Feeder lines may be introduced into the pull
system configuration, where lines feed the final assembly
line. Further, alternate operating routes for the products
along the line may be introduced.
6.4 CONWIP
CONWIP is a kanban system working with constant work-
in-process. CONWIP is a generalized form of kanban.
Like, kanban system, it relies on signals, which could be
electronic and it is equivalent to kanban cards. In a
CONWIP system, the cards traverse a circuit that includes
the entire production line. A card is attached to a standard
container of parts at the beginning of the line. When the
container is used at the end of the line, the card is removed
and sent back to the beginning of the line where it waits in a
card queue to eventually be attached to another container of
parts.
Oscar Rubiane et al. [55] have reviewed the literatures
and presented the benefits and comparison of the CONWIP
systems. Most of the articles reveal that the CONWIP
system works more efficiently than the conventional
kanban systems. Yang and Kum Khiong [96] compared 3
different systems viz., Single Kanban, Dual Kanban and
Conwip. The results show that CONWIP consistently
produces the shortest mean customer waiting time and
lowest total work-in-process. Spearman et al. [79] have
stressed that the flexibility of CONWIP system allows it to
be used by any product-line where the utility of kanban
system is limited. Hence, the superiority of CONWIP pull
system is an alternative to kanban system. They present
theoretical arguments and simulation study of CONWIP.
Christelle Duri et al. [8] have analyzed CONWIP
system, which consists of three stations in series. When a
finished part is consumed by a demand, a raw part is
released immediately and gets processed at each station
sequentially. The processing at each station does not
always meet the requirement of quality. Hence, at the end
of processing in a station, the part is checked for quality
and if it is not as per the standard, then it is sent back to the
same station for reprocessing. They proposed an analytical
method to evaluate the performance of this kind of system.
In this paper, only three stations in series are considered.
As an extension, a CONWIP system with generalized,
n stations in series may be analyzed.
6.5 SCM
There are number of articles in SCM (Supply Chain
Management). In this present survey, a few JIT-SCM
related articles are reviewed. In pull production manage-
ment systems such as JIT, deliveries must be made on an
403
as-needed basis only, and production begins only when
requested. It is supposed to match customer demand, that
is, producing only enough to replenish what the customer
has used or sold.
F. Elizabeth Vergara et al.[18] have dealt the co-
ordination between different parts of simple supply chain.
Materials should be moved from one supplier to other
supplier as per the JIT. For this, an evolutionary algorithm
was used which identifies the optimal or near optimal,
synchronized delivery cycle time and suppliers’ compo-
nent sequences for a multi-supplier, multi-component
simple supply chain. The evolutionary algorithm also
calculates a synchronized delivery cycle time for the entire
supply chain, the cumulative cost throughout the supply
chain, and the cost to each supplier. The results of this
algorithm were compared with enumeration method and
found that the evolutionary algorithm gives better solution
in quick manner. This algorithm uses only two-point
crossover genetic operators. A third genetic operator may
be introduced to further improve the performance of the
evolutionary algorithm. The evolutionary algorithm may
be modified to handle complex supply chain problem.
Stefan Minner [80] did a comprehensive review of
multiple-supplier inventory models in supply chain
management. SCM discusses strategic aspects of supplier
competition, operation flexibility, global sourcing and
inventory models. Further it was extended to logistics
and multi echelon system. The emerging importance of E-
business, especially E-procurement possibilities with the
use of Internet technologies reduces transaction costs for
supplier search and order placement with several suppliers
and therefore multiple-supplier models are more attractive
when compared to single sourcing alternative. This type of
market with spot offers, continuously changing suppliers
and high uncertainty with respect to lead-time and
reliability of supplies, makes multiple-supplier replenish-
ment and inventory strategies outperform single sourcing
policies. Matheo et al. [47] have carried out a case study on
inventory management in a multi-echelon spare parts
supply chain. This paper clearly shows the close relation-
ship between supply chain structure and demand patterns.
The problems of managing supply chain with various
numbers of echelons, multi model, extremely variable
demand and lack of visibility over the distribution channel
are discussed. They provided an algorithmic solution
through the comprehension of the sources of demand
variability and through a probabilistic forecast and inven-
tory management. Isreal David et al. [29] have enumerated
the vendor-buyer inventory production models. They argue
that there should be a certain degree of independence
between successive links of the supply chain, to allow
flexibility in production management in individual links.
They identified the degree of independence and level of
flexibility in terms of lot sizing and delivery scheduling in a
single-vendor-single-buyer system. In these lines, appro-
priate two-sided vendor-buyer inventory production mod-
els are formulated and analyzed.
In all the papers, simulation as well as meta-heuristics
can be used as powerful tools to derive results under
404
probabilistic conditions. Future research can be focused in
these lines.
6.6 Special cases
Sarah M. Rayan et al. [69] have defined POLCA system.
POLCA stands for Paired Cell Overlapping Loop of Cards
with Authorization. This system assumes that the factory
has been partitioned into non-overlapping manufacturing
cells. POLCA maintains constant WIP (like CONWIP)
between every pair of cells that experiences inter cell part
movement. Part release to a cell requires an appropriate
kanban cards as well as an authorization from the factory
loading system. CONWIP and POLCA achieve a better
trade-off between total WIP and total throughput time than
that of other systems. Their application of single chain
analysis for multiple chain operation raises an open
question whether a single WIP level should be maintained
for all products or individual levels for each product.
Further, most of the studies use simulation. Hence, future
research should be directed to develop improved search
procedures for finding WIP levels in kanban systems.
Krieg et al. [38] considered a kanban controlled
production system with 3 or more different products
processed on a single manufacturing facility as a decom-
posed system. The customers for a product arrive as per
poisson distribution. The service time and set-up changes
are product specific and follow exponential distribution. If
the customer’s demand cannot be met from stock, the
customer leaves and satisfies his demand elsewhere (lost
sales). The production run continues until the target
inventory level given by the kanbans for the product has
been reached. Then the manufacturing facility is set-up for
producing the next product. The basic principle of the
decomposed method is to decompose the original multi-
product system into a set of single-product subsystems (one
for each product). Each subsystem is modeled approxi-
mately by a continuous-time Markov chain. In this buffer
allocation problem, the objective is to minimize the
average work-in-process subject to a minimum required
throughput and a constraint on the total buffer space. The
results of the decomposition method are compared with
that of exact method and simulation. The key performance
measures are with small relative errors for the decompo-
sition method. As an extension to this work, a decompo-
sition algorithm can be developed for multi-product
kanban systems with state dependent setups. Takashashi
et al. [84] have proposed a decentralized reactive kanban
system for multi-stage production and transportation
system with unstable changes in product demand. In the
proposed system, the time series data of the demand from
the succeeding stage are monitored at each stage
individually and unstable changes in the demand are
detected by utilizing control charts. In order to develop a
control rule of the buffer size, the multi-stage production
and transportation system is decomposed into single-stage
processing systems and the performance of the decom-
posed system is investigated by simulation experiments
under various stable-demand conditions. The performance
of this system shows superior result.
Philipoom et al. [60] have done a kanban design with
flexible Toyota’s system. This system can dynamically
adjust the number of kanbans at each workstation in an
unstable production environment according to need/
demand and lead time to reduce the cost. The work of
Tardif et al. [85] introduces a new adaptive kanban-type
pull control mechanism which determines the timings to
release or reorder raw parts based on customer demands
and inventory back orders. In the adaptive pull system, the
number of kanbans in the system is dynamically readjusted
based on current inventory level and backorder level.
Unlike a conventional system, this system absorbs extra
kanbans according to the variability in demand. It was
found from the results of a simulation study of a single-
stage, single product kanban system that these systems are
beneficial in production line under variable demand
conditions. It shows that this adaptive system under such
conditions outperforms the traditional kanban pull control
mechanism. This adaptive approach may be extended for
multi-stage, multi-product kanban system.
Kutc So [40] presents the buffer allocation problem with
the objective of minimizing the average work-in-process
subject to minimum required throughput rate and constraint
on the total buffer space availability. Both the balanced and
unbalanced lines were considered in this work. On the basis
of empirical results, he developed a good heuristic for
selecting the optimal buffer allocations. The mathematical
model discussed in this paper is based on the two
assumptions that there are always materials available for
processing at the beginning of the line and that the last
station can never be blocked. But one can assume that the
first station can start processing when there are jobs
waiting which arrived as per poisson arrival pattern
(make-to-order environment). In contrast, one can consider
the situation where there is finite buffer after the last
station where the finished items are consumed by demands
with poisson pattern (make-to-stock environment). Un-
fortunately, in either case, the resulting Markov chain has
infinite number of states. So, one can develop simulation
model as the last resort for studying the problem under
these two environments.
Lai et al. [41] have proposed a system dynamic (SD)
methodology for studying the new generation of JIT in
electronic commerce environment. It is a framework for
thinking about how the operating policies of a company
and its customers, competitors and suppliers interact to
shape the company’s performance over time. The system
dynamic is a study which deals with the information feed-
back and its evolution into future decision making. It
provides a new analysis of logistics policies of the
company. Future work can be on extending the variables
and elements and to conduct experiments to investigate the
stability of the system under various conditions such as the
sudden increase in demand and random demand, experi-
mentation on the system behaviour of different types of
customer and modes of manufacturing.

The papers by Yannick Frein et al. [97] and Yves Dallery
et al. [100] introduce a new mechanism for the coordination
of multistage manufacturing system called extended kanban
control system (EKCS). It depends on two parameters per
stage viz., the number of kanbans and base stocks of finished
products. This EKCS is evolved from combining classical
kanban and base stock control system. The advantages of the
extended kanban control system were compared with the
generalized kanban control system (GKCS). It was found
that the capacity of EKCS depends only on the number of
kanbans but not on the base stock of finished parts.
A work done by Chan [6] describes the practical
approach to determine the optimal kanban size using
simulation. The research was done basically for single
product and multi product manufacturing environments in
two types of JIT production systems, the pull-type and
hybrid-type. Their measures of performance obtained
through simulation models are compared. For a single
product, when there is an increase in kanban size, the fill
rate decreases whereas with both in-process-inventory and
the manufacturing lead time increase. For multi-products
manufacture, when there is an increase in kanban size,
there is an increase in the fill rate with a decrease in the
manufacturing lead-time. Leyuan Shi and Shulimen [43]
have presented a hybrid algorithm for buffer allocation
problem called the hybrid nested partition method (NP) and
tabu search method (TS). The nested partitioned method is
globally convergent and can utilize many of the existing
heuristic methods to speed up its convergence. In this
paper, tabu-search is incorporated in the nested partitioned
framework and it was found that such incorporation results
in superior solutions. The new algorithm is efficient for
buffer allocation problems in larger production lines. The
nested partitioned method can be enhanced by incorporat-
ing any one or a combination of the many other heuristics
viz., elaborate partitioning, sampling, backtracking
scheme, simulation, etc. Then, they can be applied to
combinatorial problems of this type.
7 JIT integration, implementation and benefits
Just-in-time is a manufacturing philosophy by which an
organization seeks continuous improvements. For ensuring
continuous improvements, it is necessary for any organi-
zation to implement and integrate the JIT and JIT related
areas. If it is practiced in its true sense, the manufacturing
performance and the financial performance of the system
will definitely improve.
Swanson et al. [83] have reiterated that proper planning
is essential for implementation of a JIT manufacturing
system and a commitment from top management is a pre-
requisite. Cost benefit analysis is to be studied initially with
the knowledge of key items such as the cost of conversion
to a JIT system and time period of conversion. Cook et al.
[11], in their case study for applying JIT in the continuous
process industry, show improvements in demand forecast
and decrease in lead-time variability.
405
The relationship between implementation of TQM, TPM
and JIT will lead to improvement in the manufacturing
performance (Kribty et al. [37]). Further Huang [23]
discusses the importance of considering the integration of
TPM, JIT, Quality control and FA (Factory Automization).
Imai [27] believes that TQM and TPM are the two pillars
supporting the JIT production system. Kakuro Amasaka
[32] proposes a new JIT management system, which helps
to transfer the management technology into management
strategy.
Fullerton et al. [65] have conducted a study in 253 firms
in USA to evaluate empirically whether the degree with
which a firm implements the JIT practices affects the firms
financial performance. From their study, JIT manufacturing
system will reap sustainable rewards as measured by
improved financial performance. Also, they studied the
benefits of JIT implementation in 95 firms in USA. They
have concluded that JIT implementation improves the
performance of the system, because of resultant quality
benefits, time based benefits, employees flexibility,
accounting simplification, firms profitability and reduced
inventory level.
8 Conclusion
The growing global competition forces many companies to
reduce the costs of their inputs so that the companies can
have greater profit margin. There are considerable
advancements in technology and solution procedures in
reality, to achieve the goal of minimizing the costs of
inputs. JIT-KANBAN is an important system, which is
used in production lines of many industries to minimize
work-in-process and throughput time, and maximize line
efficiency. In this paper, the authors have made an attempt
to review the state-of-art of the research articles in the area
“JIT-KANBAN system”. After a brief introduction to push
and pull systems, different types of kanban and their
operating principles, blocking mechanisms, the authors
have classified the research articles under JIT-KANBAN
system into five major headings, viz., empirical theory,
modeling approach, variability and its effect, CONWIP and
JIT-SCM. Also, the authors have provided a section for
special cases under JIT-KANBAN. This paper would help
the researchers to update themselves about the current
directions and different issues under JIT-KANBAN
system, which would further guide them for their future
researches.
The directions for future researches are presented below.
The flow shop as well as mixed model assembly line
problems come under combinatorial category. Hence,
meta-heuristics viz., simulated annealing, genetic algo-
rithm and tabu search may be used to find solution to
determine the minimum number of kanbans and other
measures. In simulated annealing algorithm, researchers
can aim to device a better seed generation algorithm which
will ensures better starting solution. In most of the papers,
comparisons are done only based on relative improve-
ments. Instead of this approach, comparisons based on
406
complete ANOVA experiments would provide reliable
inferences.
Under batch processing system with multiple product
types, research may be directed to study the effect of
different combinations of probability distributions for
arrival process and processing times on the average
inventory level of each product as finished goods.
As an extension to the work of Markham et al. [28] on a
procedure based rule induction approach for determining
the number of kanbans and other factors in JIT, develop-
ment of knowledge acquisition for this domain will be a
significant contribution to literature.
Sarkar et al. [72] did stage-wise optimization for a multi
stage kanban system for short life-cycle product in the
market. This work may be extended for multi-product with
varying production rate at each workstation in an
assembly-type production.
As an extension of the work of Mehmet Savsar et al.
[48], a study can be carried out to compare the effects of
different combinations of the two kanban withdrawal
policies and number of kanbans between workstations, on
the performance measures. In the work of Yavuz and Satir
[98] on multi-item, multi-stage flow line, simulation model
may be developed with feeder lines introduced into the pull
system configuration, where lines feed the final assembly
line. Further, alternate operating routes for the products
along the line may be introduced.
As an extension to the work of Chaturvedi and Golhar
[7], a CONWIP system with generalized, n stations in
series may be analyzed.
This algorithm developed by Elizabeth Vergara et al.
[18] uses only two-point crossover genetic operators. A
third genetic operator may be introduced to further improve
the performance of the evolutionary algorithm. The
evolutionary algorithm may be modified to handle com-
plex supply chain problem. In JIT-SCM related research
works, effort should be directed to develop simulation as
well as meta-heuristics to derive results under probabilistic
conditions.
In the work of Sarah M. Rayan et al. [69], the application
of single chain analysis for multiple chain operation raises
an open question whether a single WIP level should be
maintained for all products or individual levels for each
product. Further, most of the studies use simulation. Hence,
future research shall be directed to develop improved
search procedures for finding WIP levels in kanban
systems. As an extension to the work of Krieg et al. [38],
a decomposition algorithm can be developed for multi-
product kanban systems with state dependent setups. The
adaptive approach suggested by Tardif et al. [85] may be
extended for multi-stage, multi-product kanban system.
The work of Lai et al. [41] can be extended by including
more variables and elements and conducting experiments
to investigate the stability of the system under various
conditions such as the sudden increase in demand and
random demand, experimenting on the system behaviour of
different types of customer and modes of manufacturing.
The nested partitioned method provided by Leyuan Shi and
Shuli Men [43] can be enhanced by incorporating any one
or a combination of the many other heuristics viz.,
elaborate partitioning, sampling, backtracking scheme,
simulation, etc. Then, they can be applied to combinatorial
problems of this type.
Ants colony optimization algorithm is a recent inclusion
to the existing meta-heuristics viz., simulated annealing
algorithm, genetic algorithm and tabu search. So, a
researcher can study the solution accuracy as well as
required computational time of this algorithm for his/her
JIT problem of interest, which falls under combinatorial
category and compare its results with the results of the
other three heuristics (meta-heuristics).
Acknowledgement The authors thank the unanimous referees for
their constructive criticisms, which helped them to improve the
content and presentation of this review paper.
References
1. Alabas C, Altiparmak F, Dengiz B (2002) A comparison of the
performance of artificial intelligence techniques for optimizing
the number of kanbans. J Oper Res Soc 53(8):907
2. Petroni A, Rizzi A (2002) A Fuzzy logic based methodology to
rank shop floor dispatching rules. Int J Prod Econ 76:99–108
3. Bitran GR, Chang (1987) A mathematical programming
approach to a deterministic Kanban system. Manage Sci
33:427–441
4. Blair, Berkley J (1992) A review of the kanban production
control research literature. Prod Oper Manag 1:393–411
5. Brown KA, Mitchell TR (1991) A comparison of just in time
and batch manufacturing the role of performance obstacles.
Acad Manage J 34(4):906–917
6. Chan FTS (2001) Effect of kanban size on just in time
manufacturing system. Journal of Material Processing Tech-
nology 116:146–160
7. Chaturvedi M, Golher DY (1992) Simulation modeling and
analysis of JIT production system. Prod Plan Control 3(1):81–92
8. Christelle D, Yannick F, Lee H-S (2000) Performance
evaluation and design of CONWIP system with inspection.
Int J Prod Econ 64:219–229
9. Panayiotou C G, Cassandras C G (1999) Optimization of
kanban - based manufacturing systems. Automatica 35c:
1521–1533
10. Co Henry C, Sharafali, Moosa (1997) Overlapping factor in
Toyota’s formula for computing the number of kanbans. IIE
Trans 29(5):409–415
11. Cook R L, Robert A (1996) Applying JIT principles to
continuous process manufacturing supply chains. Prod Invent
Manage J 37(1):12–17
12. Golhar Damodar Y, Stamm C L (1991) The just in time
philosophy: a literature review. International J Prod Res 29
(4):657–676
13. Davis WJ, Stubitz SJ (1987) Configuring a Kanban system
using discrete optimization of multiple stochastic responses. Int
J Prod Res 25:721–740
14. Deleersnyder JL, Hodgson TJ, Muller Malek H, O’Grady PJ
(1989) Kanban controlled pull system an analytic approach.
Manage Sci 35:1079–1091
15. Di Mascolo et al (1996) An analytical method for performance
evaluation of kanban controlled production systems. Oper Res
44(1):50–64
16. Drexl et al (2001) Sequencing JIT mixed-model assembly lines
under station-load and part-usage constraints. Manage Sci 47
(3):480
17. Hay E J (1988) The just in time break through. John Wiley &
sons, New York

18. Vergara EF, Moutaz K, Zhigniew M (2002) An evolutionary
algorithm for optimizing material flow in supply chain. Comput
Ind Eng 43:407–421
19. Gabrial T, Bicheno J, Galletly JE (1991) JIT manufacturing
simulation. Industrial Management & Data system 91:4
20. Gravel M, Prince W (1988) Using the Kanban in a job shop
environment. Int J Prod Res 26(6):1105–1118
21. Hemamalini B, Rajendran C (2000) Determination of the
number of containers, production kanbans and withdrawal
kanbans; and scheduling in kanban flowshop- Part I. Int J Prod
Res 38(11):2529–2548
22. Hemamalini B, Rajendran C (2000) Determination of the
number of containers, production kanbans and withdrawal
kanbans; and scheduling in kanban flowshop- Part II. Int J Prod
Res 38(11):2549–2572
23. Huang P (1991) World class manufacturing in 1990s integrat-
ing JIT,TQC,FA, and TPM with worker participation, modern
production concept theory and application. Springer, Berlin
Heidelberg New York 491–507
24. Huang PY, Rees LP, Taylor BW (1983) A simulation analysis
of Japanese Just In Time technique with Kanban for multi line
multi stage production system. Decis Sci 14:326–344
25. Wang H, Wang H-P(Ben) (1999) Optimum number of kanban
between two adjacent workstations in JIT system. Int J Prod
Econ 22:179–188
26. Hurrion RD (1997) An example of simulation optimization
using a neural network metamodel: finding the optimum
number of kanbans in a manufacturing system. J Oper Res Soc
48(11):1105–1112
27. Imai M (1988) Will America corporate theme song be Just In
Time ? J Qual Partic 21(2):26–28
28. Markham Ina S et al (1998) A rule induction approach for
determining the number of kanbans in a Just in Time
production system. Comput Ind Eng 34(4):717–727
29. Israd David, Mosh Eben C (2003) How for should JIT vendor
buyer relationship go? Int J Prod Econ 81:361–368
30. Greene James H (1987) Production and inventory control hand
book - American production & Inventory control society, Mc
Graw Hill Book Company
31. Jing-wen-Li (2003) Improving the performance of Job shop
manufacturing with demand-pull production control by redu-
cing setup/processing time variability. Int J Prod Econ 84:255–
270
32. Kakuro A (2002) New JIT A new management technology
principle at Toyota. Int J Prod Econ 80:135–144
33. Karmarker and Kekre (1989) Batching policy in kanban
system. J Manuf Syst 8:317–328
34. Kern et al (1996) Master production rescheduling policy in
capacity-constrained just- in-time make-to-stock environments.
Decis Sci 27(2):365–387
35. Khan, Lutfar R, Sarker R A (2002) An optimal batch size for a
JIT manufacturing system. Comput Ind Eng 42(2–4):127
36. Kimura O, Terade H (1981) Design and analysis of pull system
a method of multistage production control. Int J Prod Res
19:241–253
37. Cua K O, Kathleen E, Mekone, Schroeder R G (2001)
Relationship between implementation of TQM, JIT and TPM
and manufacturing performance. J Oper Manag 19:675–694
38. Krieg G, Kuhn H (2002) A decomposition method for multi
product kanban systems with setup times and lost sales. IIE
Trans 34(7):613
39. Kulwiec R A (1997) The down side of JIT. Mod Mat Handl 52
(10):3
40. Kutc S (1997) Optimal buffer allocation strategy for minimiz-
ing work in processinventory in un paced production lines. IIE
Trans 29:81–88
41. Lai CL, Lee WB, W.H.Ip (2003) A study of system dynamics
in Just in Time logistics. Journal of Material Processing
Technology 138:265–269
42. Lee CY Lin, Uzosoy CS, Wong R (1994) K., Implementation
of demand pull system in a job shop environment. Int J Prod
Res 32(11):2915–2927
407
43. Leyuan S, Shulimen (2003) Optimal buffer allocation in
production lines. IIE Trans 35:1–10
44. Lambrecht M, Segaert A (1990) Buffer stock allocation in serial
and Assembly type of production line. Int J Oper Prod Manage
10:2
45. Markey M (1996) Examining a kanban material acquisition
system. Ind Manage 38(3):22–26
46. Mason P A (1999) MRP II and kanban formula. Logist Focus
7:19–23
47. Kalchschmidt M, Giuliozotheri, Verganti R (2003) Inventory
management in a multi echelon spare parts supply chain. Int J
Prod Econ 81 & 82:397–413
48. Mehmet S, Adbullah-Al-Jawini (1995) Simulation analysis of
Just in Time production system. Int J Prod Econ 42:67–78
49. Mohanasundaram KM, Natarajan K, Viswanathkumar,
Radhakrishnan P, Rajendran C (2002) Scheduling rules for
dynamic shop that manufacture multi-level jobs. Comput Ind
Eng 44:119–131
50. Monden Y (1983) Toyota production system. Ind Eng Manage
press, - Atlanta
51. Adam N R, Will J, Bertrand M, Surkis J (1987) Priority
assignment procedure in multi level assembly jobshops. IIE
Trans 19(3):317–328
52. Nori VS, Sarker BR (1998) Optimum number of kanbans
between two adjacent stations. Prod Plan Control 9:60–65
53. Ohno K, Nakashima K, Kojima M (1995) Optimal number of
two kinds of Kanbans in a JIT production system. Int J Prod
Res 33:1387–1401
54. Ohno T (1988) Toyota production system beyond large scale
production. Productivity Press, Cambridge
55. Ovalle O R, Marquez A C (2003) Exploring the utilization of
CONWIP system for supply chain management. A comparison
with fully integrated supply chain. Int J Prod Econ 83:195–215
56. Panneerselvam R (1999) Production and operation manage-
ment. Prentice Hall of India, New Delhi
57. Patrick R, Mc Mullen, Frazier G V (2000) A simulated
annealing approach to mixed model sequencing with multiple
objectives on Just in Time line. IIE Trans 32:679–686
58. Brucker P, Heitmann S, Huricnk J (2003) Flow-Shop problems
with intermediate buffers. OR-Spectrum 25:549–574
59. Philipoom et al (1987) An investigation of the factor influencing
the number of kanbans required in implementation of the JIT
technique with Kanbans. Int J Prod Res 25(3):457–472
60. Philipoom PR, Rees LP, Taylor BW, Huang (1987) Dynami-
cally adjusting the number of kanban system in JIT production
system using estimated values of lead time. IIE Trans 19
(2):199–207
61. Rajendran C (1999) Formations and heuristics for scheduling in
kanban flow shop to minimize the sum of weighted flowtime,
weighted tardiness and weighted earliness of containers. Int J
Prod Res 37(5):1137–1158
62. Richard J, Tersine (1994) Principle of inventory and materials
management. PTR Prentice hall Englood cliffs, New Jersey
63. Schonberger R (1987) World Class Manufacturing case Book
implementing JIT & TQC. The Free Press, New York
64. Rosemary et al (2001) The production performance benefits
from JIT implementation. J Oper Manage 19:81–96
65. Rosemary R, Fullerton, Cheryl S, Mc Watters, Fawson C
(2003) An examination of the relationship between JIT and
financial performance. J Oper Manage 20:383–404
66. Bowden RO, Hall J D, Usher J M (1996) Integration of
evolutionary programming and simulation to optimize a pull
production system. Comput Ind Eng 31(1&2):217–220
67. De Smet R, Ludo G (1998) Using simulation to evaluate the
introduction of a kanban sub system within an MRP-Controlled
manufacturing environment. Int J Prod Econ 56 & 57:111–122
68. Salameh MK, Ghattas RE (2001) Optimal Just in Time buffer
inventory for regular preventive maintenance. Int J Prod Econ
74:157–161
69. Ryan S M, Choobineh F (2003) Total WIP and WIP mix for a
CONWIP controlled jobshop. IIE Trans 35:405–418
408
70. Ryan SM, Bruno B, Choobineh F F (2000) Determining
inventory levels in CONWIP controlled jobshop. IIE Trans
32:105–114
71. Sarker B R, Balan Chidambaram V (1996) Operations planning
for kanbans between two adjacent workstations. Comput Ind
Eng 31(1&2):221–224
72. Sarker Bhaba, Balan Chidambaram V (1999) Operations
planning for a multi-stage kanban system. Eur J Oper Res
112(2):284–303
73. Seki Y, Hoshino N (1999) Transient behavior of a single-stage
kanban system based on the queuing model. Int J Prod Econ 60
& 61:369–374
74. Shahabudeen P, Krishnaiah K (1999) Design of a Bi-Criteria
kanban system using Genetic Algorithm. Int J Manage syst 15
(3):257–274
75. Shahabudeen P, Gopinath R, Krishnaiah K (2002) Design of bi-
criteria kanban system using simulated annealing technique.
Comput Ind Eng 41(4):355
76. Shahabudeen P, Krishniah K, Thulasinarayanan (2003) Design
of two card dynamic kanban system using a simulated
annealing algorithm. International Journal Advanced Manufac-
turing Technology 21:754–759
77. Sharadapriyadarshini, Rajendran C (1997) Scheduling in
kanban-controlled flow shop with dual blocking mechanisms
and missing operations for part types. Int J Prod Res 35
(11):3133–3156
78. Sharadapriyadarshini, Rajendran C (1997) Heuristics for
scheduling in kanban system with dual blocking mechanisms.
Eur J Oper Res 103:439–452
79. Spearman ML, Woodruff DT, Hopp WJ (1990) CONWIP a pull
alternative to Kanban. Int J Prod Res 28:879–894
80. Stefan Minner (2003) Multiple supplier-inventory model in
supply chain management: a review. Int J Prod Econ
81&82:265–279
81. Stockton DJ, Lindley RJ (1995) Implementing kanban with
high variety / load volume manufacturing environment. Int J
Oper Prod Manage 15(7):47–59
82. Sugimori Y et al (1977) Toyota production system and kanban
system - materialization of just in time and respect for human
system. Int J Prod Res 15(6):553–564
83. Swanson C A, Lankford W M (1998) Just In Time
manufacturing. Bus Process Manag J 4(4):333
View publication stats
84. Takashashi K, Nakamura N (2002) Decentralized reactive
kanban system. Eur J Oper Res 139(2):262
85. Tardif V, Maaseidvaag L (2001) An adaptive approach to
controlling kanban systems. Eur J Oper Res 132(2):411
86. Tayfur A, Goang A S (2000) Pull type manufacturing system
with multiple product types. IIE Trans 32:115–124
87. Turbo days (N) Time since Jan 12 (1996) http://www/geocities.
com/timesquare/1848/japan21.html
88. Uday S, Karmarkar (1987) Lot-sizing and sequencing delays.
Manage Sci 33:419–423
89. Dudek Villeda R, Smith RML (1988) Increasing the production
rate of just in time production system with variable operation
times. Int J Prod Res 26(11):1747–1768
90. Vito A, Dassisti M, Okogbaa G O (1995) Approximation
approach for performance analysis production lines under a
kanban discipline. Int J Prod Econ 40:197–207
91. Womack JP et al (1991) The machine that change the world -
the story of lean production. Rawson /Harper perennial, New
York
92. Womack JP, Jones DT (1994) From the lean production to lean
enterprise. Harward Business Review 93–103
93. Woolsey R ED, Bowden R O, Hall J D, Hadley W H (1999)
Closed-form solution to kanban sizing problem. Prod Invent
Manage J 40(1):1–3
94. Xiaobo Z, Ohno K (1997) Algorithms for sequencing mixed
models on an assembly line in a JIT Production system.
Comput Ind Eng 32(1):47–56
95. Herer Yale T, Levishalom (2000) The Kanban Assignment
problem- a non integral approach. Eur J Oper Res 120:260–276
96. Yang (2000) Managing a flow line with single-Kanban, dual-
Kanban or Conwip. Prod Oper Manage 9:349
97. Yannick F, Maria D, Yves D (1995) On the design as
generalized kanban control system. Int J Oper Prod Manage
15:9
98. Yavuz IH, Satir A (1995) A kanban based simulation study of a
mixed model just in time manufacture line. Int J Prod Res 33
(4):1027–1048
99. Yoichi S, Naoto H (1999) Transient behavior of single stage
kanban system based on the queuing model. Int J Prod Econ 60
& 61:369–374
100. Yves D, Liberopoulos G (2000) Extended kanban control
system: combining kanban and base stock. IIE Trans 32:
369–386