International journal of computer integrated manufacturing , tập 23, số 5, 2010

International Journal of Computer Integrated Manufacturing
Vol. 23, No. 5, May 2010, 391–401

Prevention of resource trading fraud in manufacturing grid: a signalling games approach
Haijun Zhanga,b*, Yefa Hua and Zude Zhoua
a

School of Mechanical and Electronic Engineering, Wuhan University of Technology, Wuhan 430070, China; bDepartment of
Mechanical Engineering, University of Michigan, Ann Arbor, MI 48109-2125, USA
(Received 13 March 2009; final version received 14 January 2010)
In the manufacturing grid resource market, the information asymmetry for buyers and sellers is a common situation.
With resource information superiority, some resource service providers (RSPs, sellers) often make the pooling
equilibrium, so that resource service demanders (RSDs, buyers) cannot recognise high-quality resources owing to
imperfect information. Hence, the authors propose a signalling games approach to prevent resource trading fraud in
the manufacturing grid. In support of the architecture of resource negotiation and trading, RSDs can get more
accurate information about resource quality based on the collateral currency promised by RSPs. In this way, RSDs
can more accurately identify resource quality. This method focuses on preventing low-quality RSPs from sending
out an incorrect signal suggesting high resource quality to entice RSDs to purchase the low-quality resource.
Simulation results indicate that the game theoretical model has a reasonable and perfect Bayesian separating
equilibrium, from which RSPs do not initiatively deviate.
Keywords: manufacturing grid; singling games; the perfect Bayesian equilibrium; trading fraud

1.

Introduction

With the rapid development of advanced manufacturing technology, information technology, and the global
market, modern manufacturing enterprises are facing
sustainable, variable, and unpredictable competition.
The traditional quality and price competitive model in

manufacturing has turned into a service, technology,
and time competitive model. In this situation, enterprises cannot continue the traditional mode of ‘big and
all-inclusive projects’; they must adopt a new ‘small
and specialised’ thought processes. Modern enterprises
must emphasise professionalism and standardisation.
In the new economy, most enterprises adopt a win–win
strategy for cooperation. Thus, a common platform
must be developed – one that promotes the sharing of
manufacturing resources and service.
The manufacturing grid (MGrid) (Qui et al. 2004,
Shi et al. 2007) is a new concept and one of the
next generations of manufacturing models. It has
been proposed to meet the cooperative demand of
the manufacturing industry. It enables geographically
distributed manufacturing resources to be connected
through the internet using grid technology. With
the MGrid platform, common resources and services
(including human, equipment, material, and software)
can be shared. Enterprise competitiveness can be
enhanced in the MGrid owing to the shortened

*Corresponding author. Email:
ISSN 0951-192X print/ISSN 1362-3052 online
Ó 2010 Taylor & Francis
DOI: 10.1080/09511921003642113


development and manufacturing time and minimised
cost. The MGrid tries to achieve the sharing of
geographically distributed manufacturing resources

and services through the reconfigurable manufacturing
processes (also called ‘virtual organisations’). Therefore, enterprises in the MGrid can achieve the
goal of TQCSEF (Time, highest Quality, lowest
Cost, best Service, cleanest Environment, and greatest
Flexibility).
With the technical support of Web services
(Mockford 2004) and Grid (Foster et al. 2001), MGrid
has constructed a huge manufacturing-oriented resource and service market. Enterprises and even
individuals can acquire manufacturing resources and
services from the MGrid resource market as conveniently as they obtain water, electricity, and internet
information. MGrid users can easily pay for resources
or services (such as the material, design, or machinery
they consume) with the records and payment functions
provided by the MGrid core middleware.
Participants in the MGrid market are divided into
two categories: they are either resource service providers (RSPs) or resource service demanders (RSDs).
Resource quality is the basis of both the RSD’s
payment and the RSP’s pricing. The aim of both
RSPs and RSDs is to maximise their own profits. As
in the commodity market, the resource information
asymmetry is for RSPs and RSDs in the MGrid


392

H. Zhang et al.

resource market. With resource information superiority, some RSPs send out a confusing signal (or
inaccurate information) regarding resource quality,
which may entice RSDs to buy or use the low-quality

resource to make the pooling equilibrium. When this
happens, RSDs cannot recognise a high-quality
resource due to imperfect information, and RSPs
make an abnormal profit. This paper focuses on
preventing RSPs who have low-quality resources
from sending out untrue information. Therefore,
resource information asymmetry induces two important issues in the MGrid resource trading: (1) how to
display resource quality information for RSPs in order
to attract the attention of RSDs and (2) how RSDs can
obtain accurate information about the resource quality
from RSPs.
MGrid resource trading involves the individual
behaviours of RSPs and RSDs. In free market
economics, whether RSPs provide high-quality resources or low-quality resources is a result of the
rational behaviours of both RSPs and RSDs. The
authors adopt the signalling game theory in the information economics to analyse individual behaviours
of both RSPs and RSDs. The signalling game theory
was chosen because it focuses on decision making and
the related market equilibrium given the interaction
between decision makers.
The rest of the paper is organised as follows:
Section 2 investigates the related work about MGrid.
Section 3 describes the MGrid resource negotiation
and trading model based on perfect Bayesian Nash
equilibrium. Section 4 simulates the model and
analyses the results. Section 5 shows an industrial
case for the model. The conclusion and future work are
given in Section 6.
2. Related work
The conception, system platform, Open Grid Services

Architecture (OGSA) and Web Services Resource
Framework (WSRF) of MGrid have been investigated
by several authors (Li et al. 2007, Tao et al. 2008a,
Zhang et al. 2008). The application demands of grid
technology in e-science, e-government, e-entertainment, e-education, e-business, and manufacturing
industries have also been studied. In such MGrid
environments, RSPs and RSDs have different goals,
objectives, strategies, and supply-and-demand patterns. Moreover, both resources and end-users are
geographically distributed with different time zones.
The economic approach provides a fair basis in
successfully managing the decentralisation and heterogeneity that is present in human economies. In an
economic-based approach, the resource scheduling is
made dynamically at runtime and is driven and

directed by the end-users’ requirements. Pricing
based on user demand and resource supply is the
main driver in the competitive, economic MGrid
resource market.
Since MGrid uses the internet as a carrier for
providing remote manufacturing services, MGrid
can be used to share manufacturing resources in a
seamless manner for cooperative problem solving.
The resource price is adjusted, according to the law
of supply and demand. Price fluctuation reflects
the market’s supply-and-demand dynamics, and the
optimal resource allocation occurs at the supply-anddemand equilibrium.
The application of economic theory in computer
system resource allocation can be traced back to 1968,
when Sutherland proposed the auction mechanism for
resource allocation in the PDP-1 computer (Sutherland

et al. 1968). Since then, the economic theory has
been primarily used for solving the load balance of
computer clusters and distributed systems. Ferguson
(1996) investigated the application of general equilibrium theory and Nash equilibrium in the distributed
computer resource management. Waldspurger (1992)
designed and developed the Spawn system, which is
the market-oriented scheduling system for a group of
heterogeneous computers connected to the internet.
Bogan (1994) used the market mechanism to allocate
the central processing unit (CPU) time and proposed
the CPU leasing agreement per unit of time. Bredin
(1998) proposed a trusted third-party arbiter to
prevent fraud in transactions between mobile agents,
which is a little similar to the notary public in Section
3.3. However, this paper focuses on how to let the
collateral currency represent the quality level of
resources and/or services truly.
In recent years, studies related to the application of
economic theory in grid resource allocation have
become very popular (Buyya et al. 2002, Subramoniam
et al. 2002). Abramson (2002) studied the resource
management, scheduling, and computational economics in a grid environment and designed the grid
architecture for computational economy (GRACE).
Buyya (2009) later developed a grid application software toolkit based on Globus–Nimrod/G. Wolski
(2000, 2001) proposed that the relative value of a
resource varies with the supply-and-demand and
allocated dynamic resources using an auction model
in G-commerce project. The Popcorn project focused
on formulating computer service time as currency
(London 1998).

Computing grid researchers have carried out an indepth study on the application of economic theory in a
large-scale, dynamic system–Grid. They achieve pretty
good results that now provide the references for
MGrid.


International Journal of Computer Integrated Manufacturing
More and more researchers are becoming aware of
the importance of economics in MGrid. Wu et al.
(2005) presented a trading support manufacturing
resource sharing model similar to P2P and deployed
core services at every node responsible for aggregating
manufacturing resource information. Giovanni (2002)
investigated the economic convenience in front of
dedicated manufacturing systems depending on the
competitive market conditions. They proposed a
theoretical model that can locate market conditions
and make scope economy manufacturing systems less
profitable than dedicated manufacturing ones, and set
some general criteria to guide the entrepreneur in
making wise investment decisions regarding these
kinds of manufacturing investments. Lu et al. (2007)
defined the competitive equilibrium of Pareto optimal
in the MGrid resource optimal allocation, in order
to realise profit maximisation for RSPs and efficiency
maximisation for RSDs.
However, insufficient research has been done on
how to display resource quality information for RSDs
and how to obtain accurate information regarding
resource quality from RSPs. There are two traditional

ways of solving this issue. First is the centralised trust
system (Resnick and Zeckhauser 2002), which assumes
that a few of the entities record the history of entities
that participate in the network, and calculate and
publish the results of the creditability evaluation for
every entity. Second is the distributed trust system
(Cornelli et al. 2002), which assumes that every entity
calculates the credibility of a certain entity in the
network, according to the behaviour evaluations
provided by others. Information is generally static in
the centralised trust system, while network communication in the distributed system makes it difficult to
calculate credibility.
The MGrid environment has the following characteristics: resource distribution and sharing, selfsimilitude, dynamics and diversity, autonomist and
multiplicity of management, and highly abstract and
transparency (Tao et al. 2009). The resource management and cooperation in MGrid is more complex and
difficult than other types of distributed information systems. Therefore, the static information in the
centralised trust system and the complex communication in the distributed system are not suitable for the
MGrid environment. It needs a simple and flexible
approach for solving the above-mentioned issues for its
resource market.
The resource trading process in MGrid is essentially a game in which RSPs and RSDs evaluate each
other’s behaviour and characteristics, and then choose
the strategies that maximise their own individual
benefits. Current state-of-the-art grid technology uses
game theory for grid resource optimised allocation

393

(Riky et al. 2008, Tao et al. 2008b). The authors
introduce the game theory into the resource trading

process in MGrid and propose an MGrid resource
trading model based on the perfect Bayesian Nash
equilibrium. In this way, the authors analyse the
trading process using a dynamic model of incomplete
information. This model makes RSPs tend not to
cheat and RSDs determine the quality of the resource
according to the collateral currency promised by
RSPs. This is a new method of solving the above
issues of MGrid resource trading. The key is to prevent
RSPs from sending out an improperly high quality
signal.
3. MGrid resource negotiation and trading model
3.1. The model of MGrid resource market
Consider resource trading in the MGrid environment:
If RSDs need some manufacturing resource or service,
they search for resources and inquire about prices in
the MGrid market, according to the catalogue of
resource. RSPs can also register and maintain resource
and service information through the MGrid portal.
The information in the MGrid market is uniformly
stored and managed by the MGrid index information
server (MGIIS), as shown in Figure 1. Suppose RSPs
who provide the same type of resources but at different
levels of quality, all charge the same price P. The RSPs
know the level of quality of resources they offer (but
the RSDs do not), and they promise a collateral
currency F. If a RSD argues that the purchased
resource does not meet contract specifications, then the
RSP compensates the RSD by paying F. Therefore,
RSDs would like to purchase resources with higher F,

given a constant P.
The goal of this paper is to establish an effective
signalling game model in which RSPs are motivated to
make a promise of a collateral currency F that
accurately reflects the quality of their resource. In
signalling game theory, the quality of resource can be
regarded as the types of RSPs given by ‘nature’, and
the collateral currency F as the signalling according to
their types. RSDs infer the quality of a given resource
from F and choose one RSP as their supplier. Then,
the dynamic signalling game model of incomplete
information would be established in order to solve the
above-mentioned issues in MGrid resource trading.
3.2.

Signalling games

Signalling games (Myerson 1997) are dynamic games
in which information transfer is viewed as the signal.
They are incomplete information games with two
players, a sender and a receiver. The type of sender (i.e.
the quality of resources that RSPs provide) is private


394

Figure 1.

H. Zhang et al.

The MGrid resource market model.

information; the type of receiver (i.e. RSDs prefer to
buy resources with a higher collateral currency, given a
constant price) is public information. The two players
receive payoffs depending on the type of sender, the
message chosen by the sender, and the action chosen
by the receiver. The sequence of signalling games is as
follows:
(1) The sender has a certain type y 2 Y, which is
given by nature. Y ¼ {y1, . . . , yK} is the set of
sender types. The sender knows its own y while
the receiver does not know the type of the
sender. The receiver only knows the prior
probability p ¼ p(y) where the type of the
sender is y, Skp(yk) ¼ 1.
(2) Based on his knowledge of his own type, the
sender chooses to send a message (m) from a set
of possible messages M ¼ {m1, . . . , mJ}.
(3) The receiver observes the message (m) but not
the type of the sender (y), and gets the posterior
probability p~ ¼ p~ (y j m) from the prior probability p ¼ p (y) according to the Bayes rule.
Then, the receiver chooses an action a 2 A
from a set of feasible actions A ¼ {a1, . . . , aH}.
(4) The payoff functions of the sender and the
receiver are u1(m, a, y) and u2(m, a, y),
respectively.
The equilibrium concept that is relevant for
signalling games is Perfect Bayesian equilibrium.
Perfect Bayesian equilibrium is a refinement of

Bayesian Nash equilibrium, which is an extension of
Nash equilibrium of incomplete information games.
Perfect Bayesian equilibrium is the equilibrium

concept relevant for dynamic games of incomplete
information.
3.3. Architecture of MGrid resource negotiation and
trading
For MGrid resource trading, an architecture for
MGrid resource negotiation and trading is designed
in the section. The architecture is sufficiently general to
accommodate different economic models used for
resource trading and service access cost determination
(see Figure 2 for more details). This consists of four
parts: RSDs, grid brokers, core grid middleware, and
RSPs. The features of GRACE (Abramson et al. 2002)
have been extended to focus on preventing resource
trading fraud (such as the credit management module).
On the basis of the resource negotiation and
trading architecture, the resource trading process is
as follows (shown in Figure 3):
(1) RSPs and RSDs login to MGIIS.
(2) RSPs and RSDs deposit money into the Grid
Bank. In this study, the authors use a ‘prepaid’
payment mechanism.
(3) RSPs publish their resource information (i.e.
process capability, price, duration, and collateral currency) in the MGrid market through the
MGrid portal to attract RSDs. RSDs can also
search and check resource information through
the portal.

(4) RSPs sign a contract with RSDs. In the
meantime, the contract must be legally notarised and the notary public takes over both
deposits in the Grid Bank.


395

International Journal of Computer Integrated Manufacturing

Figure 2.

The architecture of resource negotiation and trading in MGrid.

(5) If a trade is successful, the RSD’s deposit will
be transferred into the RSPs account by the
notary public. Otherwise, the RSP’s deposit is
transferred into the RSD’s account as compensation. No matter what, the credit system
records this trading information about RSDs
and RSPs for later credit inquiry.

3.4.

Model analysis

According to Taguchi et al. (2005), quality implies low
failure rate, low energy consumption, long service life,
high efficiency, and low damage to users. For
convenience, assume the quality level of the resource
is q (0 q 1), where higher q indicates higher
resource quality and q ¼ 1 means that the resource

has a zero failure rate. The authors employ the exponential distribution for the failure rate G(q) ¼ e–kq,
where k 2 Rþ such that k 4 0. RSPs charge P for
each resource and promise a collateral currency F. The
cost of a resource is given by C(q) ¼ a1e–b1q þ a2e–b2q
(Juran et al. 1999), where a1 refers to the loss of
defective resource as q ! 0, a2 refers to the cost of the
resource as q ! 1, and b1, b2 are the coefficients of the
function. From a practical perspective, P ! C(q) 4 0,
F 4 0. The action of RSDs is to determine the trading
probability Prob(Á). In general, RSDs consider two
factors: quality and price (that is, the cost performance). At the same P, the higher F that RSPs
promise, the higher qualities of resource RSDs assume.
Therefore, RSDs prefer to trade with RSPs who have

> 0. Therefore, the
the highest F, such that dProbðFÞ
dF
formulae of Prob(Á) is given by
ProbðFÞ ¼ m

qeðFÞ
þ n;
P

ð1Þ

where 0 5 Prob(F) 1, m 4 0,0 5 n 1, m and n
are the adjustment coefficients. q~(F) is the estimation of
resource quality after the RSD observed compensation
price F.

The MGrid resource trading model based on
perfect Bayesian Nash equilibrium is described as
follows:
(I) ‘Nature’ chooses one resource quality q
according to a certain prior probability
density, and informed RSPs know it as their
types.
(II) RSPs fix F.
(III) RSDs observe F without knowledge of q, and
then determine the trading probability
Prob(F).
(IV) The payoff function of RSPs is U (q, F, Prob
(F)).
Define the benefit as the added value of profit, not
considering the funding rate. If there is no trading, the
benefit is zero. Suppose that RSPs are risk-neutral;
therefore, the expected benefit of RSPs is
Uðq; F; ProbðFÞÞ ¼ ½ðP À F À CðqÞÞ Á GðqÞ
þ ðP À CðqÞÞ Á ð1 À GðqÞފ Á ProbðFÞ


396

H. Zhang et al.

Figure 3.

The flow chart of resource trading in MGrid.

Simplify it and get the following function:

Uðq; F; ProbðFÞÞ ¼ ½P À F Á GðqÞ À Cðqފ Á ProbðFÞ
Furthermore, the price F should satisfy U(q,F,
Prob(F)) ! 0, so the constraint of F can be obtained as
follows:
F

P À CðqÞ
GðqÞ

ð2Þ

3.5. Model solution
As a pure strategies equilibrium, the above model may
have the solutions of a pooling equilibrium, either
separating equilibrium or semi-separating equilibrium.
The goal of this study is to get the solution of

separating equilibrium, that is, the higher quality of
resource, the more likely RSPs are to promise a higher
compensation price.
Based on perfect Bayesian Nash equilibrium,
RSPs and RSDs plan optimal reactions in (II) and
(III), against the possible strategies of their opponents for each potential type of their own. The
reaction of a RSP depends on their type. Hence,
an RSP’s reaction reveals some information about
their type. RSDs can infer their opponent’s type
or revise the prior probability, and then choose
the optimal reaction. RSPs know that their reactions will be known or utilised by RSDs, and therefore try to choose reactions that most benefit
themselves.
Suppose that there exists the partial derivative of U

to F, and let @U
@F ¼ 0,


International Journal of Computer Integrated Manufacturing
@U
¼ ½P À F Á GðqÞ À Cðqފ
@F


dProbðFÞ
Â
À GðqÞ Á ProbðFÞ ¼ 0
dF

ð3Þ

If RSPs promise the collateral currency F for their
resource, u˜(q j F) is the posterior probability of q after
RSDs observe the price F. Then the expected
value of
R1
the resource quality for RSDs is qeðFÞ ¼ 0 q~
uðq j FÞdq.
u˜(q j F) can be calculated according to the Bayes
formula – a theorem that is valid in all common
interpretations of probability – in the following way:
u~ðq j FÞ ¼ R 1PðF j qÞuðqÞ
0

PðF j qÞuðqÞdq

Regarding the information set F observed by
RSDs,
the belief of RSDs 7 u˜(q j F) satisfies
R1
e
u
ðq
j
FÞdq
¼ 1.
0
For separating equilibrium, RSDs can infer q from
F because F(q) is the optimal reaction of RSPs who
provide a resource of quality q. So there is
de
qðFðqÞÞ=dq ¼ 1.
In fact, there are P(q j F(q)) ¼ 1and P(q0 j F(q))
0 for the separating equilibrium,where
q0 6¼Rq.
R1 0
1 0
0
0
Therefore,
R 1 0 0qeðFÞ ¼ 0 q ueðq j FÞdq ¼ 0 q Pðq j
0
FðqÞÞdq ¼ 0 q dðq À qÞdq ¼ q, where d is the
Kronecker function.



de
q dFðqÞ
de
q
dFðqÞ À1
Á
¼ 1;
¼
ð4Þ
dF dq
dF
dq
By combining Equations (1), (3), and (4), the
following formula can be obtained:
@U m½P À F Á GðqÞ À Cðqފ
¼
Á
@F 
À1 P
 q

dFðqÞ
À GðqÞ Á m þ n ¼ 0
dq
P

397

Next, the RSP’s strategy is analysed (as shown in
Figure 4). The curve L1 represents the inequality (2)
and the curve L2 represents the Equation (3). The two
curves intersect at M (q0,F3). When the quality of the
resource provided by a RSP (q) is less than q0, the
expected profit U of the RSP is less than zero at the
optimal compensation price F*. Therefore, the authors
do not consider there are resources with the quality
q 5 q0. In Figure 4, F1 represents the maximum of the
collateral currency which satisfies the inequality (2); F2
represents the optimal value of the compensate price
that satisfies Equation (3). RSPs with resource quality
q can gain the maximum profit by promising a
compensation price F2. The curve L2 in Figure 4
implies that F* should range from F3 to F4. Therefore,
prob(F) ¼ 0 when F 5 F3, while Prob(F) ¼ 1 when
F 4 F4. Therefore, the trading probability is given by
8
F > F4
< 1;
ProbðFÞ ¼ m eqðPFÞ þ n;
F3 F F4
:
0;
F < F3
3.6.

Model analysis

(1) The model identifies users as belonging to one of

two parties (either RSPs or RSDs) in the MGrid
resource market. The strategy of all RSDs is the
same, and that of all RSPs is also the same.
(2) The model assumes that RSPs with different
quality resources all charge the same price,
because RSDs would have a better idea of a
particular resource’s level of quality if they
were able to compare RSPs’ prices.
(3) For convenience of analysis, the authors
assume the quantity of the trading resource is
one in the model. However, the model is the
same for other quantities.

By combining C(q) and G(q) with the above
formula and solving the differential equation, the
expected value F is obtained.

P kq
a1
FÃ ¼ yðqÞ ¼ m
e À
eqðkÀb1 Þ
k
k À0
b1
a2
eqðkþb2 Þ þ C1
À
ðmq þ nPÞ
ð5Þ

k þ b2
where C1 is the integration constant. There is a game
corresponding to C1.
Let q~(F) ¼ q ¼ y–1(F*), under the observation of
rational expectations (two parties can identify the type
accurately), the payoff function of RSPs is
Uðq; F; ProbðFÞÞ ¼ ½P À F Á GðqÞ À CðqފÁ
Â

Ã
m Á yÀ1 ðFÃ Þ P þ n

Figure 4.
curves.

The compensation price & resource quality


398

H. Zhang et al.

(4) The model takes into account the cost of a
resource, given different actual qualities, while
the literature (Li et al. 2005) does not. Because
the different quality resources have different
costs, obviously, the profit is different at the
same price.
4.

Model simulation

The authors employ MATLAB V7.6.0 as the simulation tool and personal computer (Intel Pentium 4 CPU
3.40 GHz and 2GB memory, operation system: Windows Vista Enterprise) as the simulation platform.
4.1.

Simulation parameter set

A machine plant (RSD) wants to purchase a certain
part, for which all machine tool plants (RSPs) charge
P ¼ 13.50. However, the quality of the parts provided
by different machine tool plants may vary. So, the
machine plant plays games with machine tool plants.
In the MGrid resource market, machine tool plants
promise different compensation prices based on
the quality of their part. The other parameters are
C1 ¼ 1.00, m ¼ 13.095; n ¼ 0.03. According to the
historical data of products, a1 ¼ 2.64, b1 ¼ 0.98,
a2 ¼ 0.31, b2 ¼ 3.35, k ¼ 5.00 can be obtained by
means of statistical computing. The cost of a part is
given by C(q) ¼ 2.64e–0.98q þ 0.31e3.35q and the failure
rate is given by G(q) ¼ e–5q. According to the above
resource trading model, the expected profit distribution for the different resource qualities and the
different compensation prices is shown in Figure 5.
In order to validate the model, q is set to be 0.38,
0.68, and 0.98, respectively. Figure 6 shows the 36

Figure 5.

The expected profit distribution for RSPs.

expected profits of machine tool plants at the collateral
currency F for every two dollars on the closed interval
[0,F1], when the quality of resource provided by the
machine tool plant is 0.38. The machine tool plants can
obtain the maximum profit (Umax ¼ 2.15) at the
collateral currency F2 ¼ 34.59, while q is 0.38. Figure
7 shows the 28 expected profits at the collateral
currency F for every ten dollars on the closed interval
[0,F1], when q is 0.68. The machine tool plants achieve
the maximum profit (Umax ¼ 4.35) at the collateral
currency F2 ¼ 84.82 for q ¼ 0.68. Figure 8 shows the
29 expected profits at the collateral currency F for
every ten dollars on the closed interval [0,F1] for
q ¼ 0.98. The machine tool plants achieve the maximum profit (Umax ¼ 2.73) at the collateral currency
F2 ¼ 193.89 for q ¼ 0.98.

Figure 6.
q ¼ 0.38.

The expected profit distribution of RSPs at


International Journal of Computer Integrated Manufacturing
4.2.

Analysis of simulation results

(1) When the collateral currency F promised by a
RSP is the optimal value F2 corresponding to q,

the RSP achieves the maximal profit.
(2) The optimum value F2 increases with the
resource quality, and RSPs always pursue their
own individual profit maximisation in signalling games. RSDs should be able to infer the
resource quality q from F correctly.
(3) When F is greater than F1 or less than F3, the
expected profit is less than zero.

Figure 7.
q ¼ 0.68.

Figure 9.

The expected profit distribution of RSPs at

The RSOS system of MBRSSP-MGrid.

399

These results are feasible. No one would like to
purchase a resource that has a very low collateral
currency. Conversely, RSPs may take risks by paying
RSDs more collateral currency in the case of resource
breakdown if the collateral currency is too high.
The simulation results indicate that the game
theoretical model has a reasonable and perfect Bayesian
separating equilibrium, from which RSPs would not
initiatively deviate. This means that in pursuit of
maximal profits, RSPs prefer to provide RSDs with
accurate information about their resource’s quality (i.e.

the compensation price) initially and faithfully, given
the strategy of RSDs (i.e. the trading probability) in the

Figure 8.
q ¼ 0.98.

The expected profit distribution of RSPs at


400

H. Zhang et al.

signalling games model. If RSPs do not choose the
optimum value F2 as their collateral currency, then they
do not achieve the maximal profit in resource trading.
Therefore, RSPs prefer to choose a collateral currency
according their actual resource quality as opposed to
choosing an incorrect collateral currency that is higher
or lower than F2 hoping to cheat RSDs. Then, RSDs
can identify the resource quality according to the
collateral currency F2 because a higher F2 implies a
higher q.
5. Industrial case study
This methodology is employed in the development
and production of magnetic bearing. Magnetic bearing is a new type of high–performance bearing.
Compared with ordinary bearings, magnetic bearings
have lots of unique advantages, such as no mechanical contact, no friction, no wearing, no lubrication,
and a long lifetime. However, each type of magnetic
bearing must be designed and manufactured, according to the concrete applications. A large number of

resources are needed in the development of magnetic
bearings.
In order to realise the sharing and collaboration
of resources, the magnetic bearing resource and
service sharing platform in MGrid (MBRSSP-Grid)
was developed. The feasibility for the implementation
of the model is demonstrated in this experimental
platform. MBRSSP-MGrid is implemented on Globus Toolkit 4.0. Resources and services are encapsulated in the form of Web Services Resource
Framework (WSRF). RSPs can publish the information of their valued resources and services (such as
equipment, software, human, application, and technique) through the publication system of MBRSSPMGrid. Users can also search and schedule resources
or services by inquiring through the resource and
service optimal-selection (RSOS) system of MBRSSPMGrid. Figure 9 is a screenshot of the man-machine
interface of the RSOS system. RSOS provides users
with the list of services sorted by the highest collateral
currency first.
6. Conclusions and future work
In support of the architecture of resource negotiation
and trading, the authors propose an MGrid resource
trading model based on perfect Bayesian Nash
equilibrium in order to prevent trading fraud. The
model is essentially a dynamic game of incomplete
information, which best conforms to the realities of
resource trading. The theoretical analysis and simulation results indicate that the model is effective in
preventing low-quality RSPs from sending out

incorrect signals regarding their resource quality
(which may entice RSDs to buy or use the low-quality
resources). Since through this model RSDs obtain
accurate information about resources, a recommended
future work includes the combinatorial optimisation of

resource service in the MGrid, given one or more
criterion, such as minimised time, minimised cost, the
resource utilisation, and so forth.
Acknowledgements
This paper is supported by the National Natural Science
Foundation Key Project of China: Digit manufacturing basic
theories and key techniques under network environment
(NO.50335020), and the Hubei Digital Manufacturing Key
Laboratory Opening Fund project: Research on resource
service search and optimal-selection theories and experiments
in manufacturing grid system (No. SZ0621). The authors
thank the editor and the anonymous reviewers for their
constructive comments and suggestions which helped to
improve the paper.

References
Abramson, D., Buyya, R., and Giddy, J., 2002. A computational economy for grid computing and its implementation in the Nimrod-G resource broker. Future Generation
Computer Systems, 18 (8), 1061–1074.
Bredin, J., Kotz, D., and Rus, D., 1998. Market-based
resource control for mobile agents. In: 2nd International
Conference on Autonomous Agents, 9–13 May, Minneapolis, MN. New York: ACM Press, 197–204.
Bogan, N.R., 1994. Economic allocation of computation time
with computation markets. Thesis (MS). Massachusetts
Institute of Technology.
Buyya, R., Abramson, D., Giddy, J., and Stockinger, H.,
2002. Economic models for resource management and
scheduling in Grid computing. Concurrency and computation: Practice and Experience, 14 (13–15), 1507–1542.
Buyya, R., Abramson, D., and Giddy, J., 2009. The Nimrod/
G grid resource broker for economic- based scheduling,
market oriented grid and utility computing. New Jersey:

Wiley Press.
Cornelli, F., Damiani, E., Vimercati, S.D., Paraboschi, S.,
and Sanmarati, P., 2002. Choosing reputable servants
in a P2P network. In: D Lassner, ed. 11th International
WWW Conference, 7–11 May, Hawaii, USA. New York:
ACM Press, 376–386.
Ferguson, D.F., Nikolaou, C., Sairamesh, J., and Yemini,
Y., 1996. Economic models for allocating resources in
computer systems. Market-based Control: A Paradigm
for Distributed Resource Allocation, 156–183.
Foster, I., Kesselman, C., and Tuecke, S., 2001. The anatomy
of the grid: enabling scalable virtual organizations.
International Journal of High Performance Computing
Applications, 15 (3), 200–222.
Giovanni, P., Michele, A., Giovanna, L.N., and Sergio,
N.L.D., 2002. Long term capacity decisions in uncertain
markets for advanced manufacturing systems incorporating scope economies. European Journal of Operational
Research, 143 (1), 125–137.
Juran, J.M., Godfrey, A.B., Hoogstoel, R.E., and Schilling,
E.G., 1999. Juran’s quality handbook. 5th ed. New York:
McGraw Hill.


International Journal of Computer Integrated Manufacturing
Li, M.S., Yang, S.B., Fu, Q.F., and Jin, Y., 2005. Research
on grid resource reliability model based on promise. In:
International Conference on Information Technology:
Coding and Computing, 4–6 April 2005, Las Vegas, NV.
New Jersey: Institute of Electrical and Electronics
Engineers Inc., 310–339.

Li, Z., Jin, X.L., Cao, Y., Zhang, X.Y., and Li, Y.Y., 2007.
Conception and implementation of a collaborative
manufacturing grid. International Journal of Advanced
Manufacturing Technology, 34 (11–12), 1224–1235.
London, S., 1998. POPCORN – a paradigm for globalcomputing. Thesis (M.S.). The Hebrew University of
Jerusalem.
Lu, S., Zhou, Y.L., Cao, X.L., and Liu, T., 2007. Study on
economic architecture for resource management for MG
based on beneficial drive. In: International Conference on
Wireless Communications, Networking and Mobile Computing, 21–25 September 2007, Shanghai, China. New
Jersey: Institute of Electrical and Electronics Engineers
Inc., 3411–3414.
Mockford, K., 2004. Web services architecture. BT Technology Journal, 22 (1), 19–26.
Myerson, R.B., 1997. Game theory: analysis of conflict.
Cambridge, MS: Harvard University Press.
Qiu, R.G., 2004. Manufacturing grid: a next generation
manufacturing model. In: IEEE International Conference
on Systems, Man and Cybernetics, 10–13 October, The
Hague, Netherlands. New Jersey: Institute of Electrical
and Electronics Engineers Inc., 4667–4672.
Resnick, P. and Zeckhauser, R., 2002. Trust among strangers
in internet transactions: empirical analysis of eBay’s
reputation system. In: M.R. Baye, ed. The economics of
the internet and e-commerce. Bingley, UK: Emerald
Group Pub Ltd., 127–157.
Riky, S., Zomaya, A.Y., and Bjorn, L., 2008. A cooperative
game framework for QoS guided job allocation schemes
in grids. IEEE Transactions on Computers, 57 (10), 1413–
1422.
Shi, S.Y., et al., 2007. An implementation of modelling

resource in a manufacturing grid for resource sharing.
International Journal of Computer Integrated Manufacturing, 20 (2–3), 169–177.
Subramoniam, K., Maheswaran, M., and Toulouse, M.,
2002. Towards a micro-economic resource allocation in
grid computing systems. In: IEEE conference on electrical
and computer engineering, 12–15 May, Winnipeg, Manitoba, Canada, New Jersey: Institute of Electrical and
Electronics Engineers Inc., 782–785.

401

Sutherland, E., 1968. A futures market in computer time.
Communications of the ACM, 11 (6), 449–451.
Tao, F., Hu, Y.F., and Zhou, Z.D., 2008a. Study on
manufacturing grid and its executing platform. International Journal of Manufacturing Technology and Management, 14 (1–2), 35–51.
Tao, F., Hu, Y.F., and Zhou, Z.D., 2008b. Resource service
composition and its optimal-selection based on particle
swarm optimization in manufacturing grid system. IEEE
Transactions on Industrial Informatics, 4 (4), 315–327.
Tao, F., Hu, Y.F., and Zhou, Z.D., 2009. Application and
modeling of resource service trust-QoS evaluation in
manufacturing grid system. International Journal of
Production Research, 47 (6), 1521–1550.
Taguchi, G., Chowdhury, S., and Wu, Y., 2005. Taguchi’s
quality engineering handbook. New Jersey: Wiley Press.
Waldspurger, C.A., Hogg, T., Huberman, B.A., Kephart,
J.O., and Stornetta, W.S., 1992. Spawn: A distributed
computational economy. IEEE Transactions on Software
Engineering, 18 (2), 103–117.
Wolski, R., Plank, J.S., Brevik, J., and Bryan, T., 2000.
Analyzing market-based resource allocation strategies

for the computational grid. International Journal of High
Performance Computing Applications, 15 (3), 258–281.
Wolski, R., Plank, J.S., Brevik, J., and Bryan, T., 2001. Gcommerce: market formulations controlling resource
allocation on the computational. In: 15th international
parallel and distributed processing symposium, 23–27
April, San Francisco, CA. New Jersey: Institute of
Electrical and Electronics Engineers Inc., 262–269.
Wu, L., Meng, X.X., Liu, S.J., Tong, Y.X., and Wei, M.,
2005. A trading supported manufacturing resource
sharing model for manufacturing grid. In: 9th international conference on computer supported cooperative work
in design, 24–26 May, Coventry, United Kingdom, New
Jersey: Institute of Electrical and Electronics Engineers
Inc., 339–344.
Zhang, H.J., Hu, Y.F., and Zhou, Z.D., 2008. Study on
semantic-aware manufacturing grid architecture. In: 5th
international conference on fuzzy systems and knowledge
dsicovery, 18–20 October, Shangdong, China. New
Jersey: Institute of Electrical and Electronics Engineers
Inc., 626–630.


International Journal of Computer Integrated Manufacturing
Vol. 23, No. 5, May 2010, 402–411

Integrated line balancing to attain Shojinka in a multiple straight line facility
Hadi Go¨kc¸ena, Yakup Karab* and Yakup Atasagunb
a

Department of Industrial Engineering, Faculty of Engineering and Architecture, Gazi University, Maltepe, 06570, Ankara, Turkey;
Department of Industrial Engineering, Faculty of Engineering and Architecture, Selc¸uk University, Campus, 42031, Konya, Turkey

b

(Received 3 January 2009; final version received 14 January 2010)
Traditional straight assembly lines are still one of the most important elements and an important fact of today’s
production systems. If applicable, a company can combine its multiple straight assembly lines and obtain many
advantages of Shojinka more or less. This paper analyses a new problem – integrated balancing of multiple straight
assembly lines (MSLB) to attain Shojinka in a multiple straight assembly line facility. The MSLB problem is built on
the concept that it could be possible for a company to obtain the advantages of Shojinka even if the company has
not adopted the U-shaped line layout. Three connectivity types are suggested to integrate multiple assembly lines.
A binary integer formulation for integrated balancing of multiple assembly lines is developed. The objective of the
proposed formulation is to minimise the total number of workstations required in the assembly facility. The
formulation is explained and validated using some illustrative examples. The proposed approach provides flexibility
to minimise the total idle times of the lines and total number of workstations that are required in the assembly line
facility.
Keywords: assembly line balancing; integer programming; Shojinka

1.

Introduction

‘Shojinka’ is a Japanese word that is a combination of
sho (to reduce), jin (worker) and ka (to change)
(Sennott et al. 2006). The concept of Shojinka, which
was originally an important element of Toyota
Production System (TPS), is easily to increase or
decrease the number of workers in a production facility
when the demand rate is increased or decreased. In a
production facility, different types of products may be
produced on different lines. The fluctuations in

demands of products will probably require adding
workers to some lines and removing from others.
Attaining flexibility in the number of workers at a
workshop to adapt to demand changes is called
Shojinka, which is equivalent to increasing productivity by the adjusting and rescheduling of human
resources (Monden 1993). Shojinka can be attained
by changing the number of operations assigned to a
worker, who is capable of performing multiple operations. On the other hand, the machinery layout must
be proper to enable workers easily to walk between
machines. The TPS adopts U-shaped machinery layout
to form production lines. The U-shaped layout has
several advantages over other layout types such as bird
cages, isolated islands and linear (straight) layouts
(Monden 1993). The visibility and communication
between workers on U-shaped production lines are

*Corresponding author. Email:
ISSN 0951-192X print/ISSN 1362-3052 online
Ó 2010 Taylor & Francis
DOI: 10.1080/09511921003642162


strengthened. In addition, the number of workers
required on a U-shaped line will be less than or equal
to the number of workers required on a comparable
straight line (Miltenburg and Wijngaard 1994). For a
given cycle time, the number of workers required on a
U-shaped line may be fractional such as 3.4 workers.
Four workers must be assigned to this U-shaped line to
meet the demand. Hence, 60% of available time for

one of the workers will be idle (waste) on this line. That
is, (4.0–3.4)/4.0 ¼ 15% of total available workforce
is unproductive. In order to make this idle time
productive, several U-shaped lines are combined into
one integrated line by locating them close to each
other. This way, a worker can perform operations
from two or more neighbour U-shaped lines, and idle
times can be eliminated or reduced. Such a production
facility that consists of a group of U-shaped lines
is defined as a Just-in-Time (JIT) production unit
(Sparling 1998).
The problem of assigning operations to workstations on U-shaped lines, the single U-line balancing (SULB), was first studied by Miltenburg and
Wijngaard (1994). Essentially, the SULB problem is
the U-shaped line version of a well-known problem,
the simple assembly line balancing (SALB), which was
introduced by Salveson (1955). Although numerous
research papers on SALB have been published, the


International Journal of Computer Integrated Manufacturing
literature on SULB is relatively small. It is not
practical to present the literature on SALB here.
Nevertheless, several review and assessment papers can
be useful for interested readers (Baybars 1986, Ghosh
and Gagnon 1989, Erel and Sarin 1998, Becker and
Scholl 2006). Following the pioneering study of
Miltenburg and Wijngaard (1994), many researchers
attempted to develop SULB procedures (Urban 1998,
Scholl and Klein 1999, Erel et al. 2001, Aase et al.
2004, Go¨kc¸en and A

gpak 2006). Since the U-shaped
lines are a consequence of Shojinka, their advantages
can be exactly obtained when they are combined in
comparison to their independent cases. Balancing
combined multiple U-shaped lines (or JIT production
units) was studied by only Sparling (1998), Miltenburg
(1998) and Chiang et al. (2007). These studies refer to
the same problem by different names as the N U-line
balancing problem with travel (NULB-T) by Sparling
(1998), the U-line facility problem (ULF) by Miltenburg (1998) and multiple U-line balancing problem
(MULB) by Chiang et al. (2007).
The U-shaped production lines mentioned and
emphasised by Monden (1993), Sparling (1998),
Miltenburg (1998) and Chiang et al. (2007) mostly
consist of manufacturing operations which need
machinery. But, this does not mean that the U-shaped
lines cannot be used for assembly operations. Furthermore, most SULB studies emphasise assembly operations. On the other hand, it could be possible for a
company to obtain the advantages of Shojinka even if
the company has not adopted the U-shaped line
layout. It should be noted that traditional straight
assembly lines are still one of the most important
elements and an important facet of today’s production
systems. Therefore, if applicable, a company can
combine its multiple straight assembly lines and obtain
many advantages of Shojinka more or less. In a
multiple assembly line facility, the parent-component
relationships between the items of a final product
designate the location of multiple assembly lines.
Depending on the structure of the final product and
production policies, a company may have one or more

sub-assembly line. In many industries such as automotive, aircraft, electronics, machinery, etc., companies usually assemble subassemblies and final products
in their own production systems and purchase parts
and components from outside vendors. If multiple
assembly lines are close to each other then they can be
balanced simultaneously using some connectivity conditions between these lines. Balancing multiple assembly lines simultaneously is originally an integration of
assembly lines. This requires integrated balancing of
multiple lines and provides an opportunity to reduce
the number of workstations utilised in such production
systems.

403

The literature on balancing straight assembly lines
includes examples of combining multiple straight
assembly lines. Go¨kc¸en et al. (2006) suggested that in
a production facility, two or more straight assembly
lines can be located in parallel and they can be
balanced simultaneously. They developed a binary
formulation and proposed a heuristic procedure for
balancing of parallel straight assembly lines with the
objective of minimising the number of workstations
required in the system. Balancing parallel assembly
lines simultaneously will constitute workstations containing tasks on two adjacent and parallel lines and
these common (multi-line) workstations will provide
the flexibility to minimise the total idle times of the
lines and total number of workstations required in
the production facility. Go¨kc¸en et al.’s (2006) study
assumes that multiple assembly lines are located in
parallel and parallel connections can be constructed
between two lines. Essentially, Go¨kc¸en et al.’s (2006)

study can be considered as a partial implementation
of Shojinka in a parallel assembly line facility. Go¨kc¸en
et al. (2006) provide a framework that combines two
parallel assembly lines with common workstations.
However, in practice, two or more assembly lines can
be combined with common workstations using several
connectivity opportunities. But, there is no study that
combines straight assembly lines to attain Shojinka in
a multiple straight line facility. This is the first study
that proposes an assembly line balancing procedure to
attain the benefits of Shojinka in a multiple straight
line facility. This paper analyses a new problem, which
is the integrated balancing of multiple straight
assembly lines (MSLB). The objective of the problem
is to assign tasks to a minimum number of workstations on multiple straight assembly lines. Possible
location forms and connectivity opportunities among
straight assembly lines are investigated and given in
Section 2. Based on the connectivity types detected, a
binary formulation for MSLB is presented in Section 3.
The objective of the proposed formulation is to
minimise the number of workstations required in a
multiple straight assembly line facility. The proposed
formulation is illustrated using examples in Section 4.
Some concluding remarks and opportunities for
further research are presented in Section 5.
2.

Combining multiple straight assembly lines

Two assembly lines can be combined by connecting

them with one or more common workstations. Thus,
operators can work in two or more different assembly
lines at the same time. These connections may provide
an opportunity to reduce the total labour requirement
of the system. In this section, we propose that multiple
straight assembly lines can be combined using several


404

H. Go¨kc¸en et al.

connectivity opportunities between these lines. Go¨kc¸en
et al. (2006) suggested that two or more straight
assembly lines can be located in parallel to each other
and they can be balanced simultaneously. Balancing
parallel assembly lines may result in common workstations that include tasks from two adjacent lines.
These common workstations will provide flexibility to
minimise the total idle times of lines and total number
of workstations required in the production system.
Go¨kc¸en et al.’s (2006) approach is based on the
concept of parallel connectivity of two adjacent
assembly lines located in parallel. Figure 1 shows two
simultaneously balanced parallel assembly lines with
parallel connectivity.
It can be shown in Figure 1 that the parallel
connectivity appears in two common workstations.
The first common workstation includes task 3 of line 1
and tasks 2 and 3 of line 2. The second common
workstation includes tasks 4 and 6 of line 1 and task 5

of line 2. Parallel connectivity can be benefited when
two or more assembly lines are located in parallel and
they are sufficiently close to each other. However, it
should be noted here that a common workstation
includes tasks from only two adjacent lines. This is due
to the difficulty and inefficiency for operators to travel
from one line to another non-adjacent line.
In a production system, it may not be always
possible to locate all assembly lines in parallel.
Furthermore, if assembly lines are tightly related to
each other and some supplier–customer relationships
exist between these assembly lines then they may be
located in different ways. In addition to the parallel
connectivity, we suggest two new connectivity types,
which are consecutive and perpendicular connectivity. It
should be noted here that one or more common
workstations are required to obtain a parallel

Figure 1.

A parallel connectivity.

Figure 2.

A consecutive connectivity.

connectivity between two assembly lines. However,
only one common workstation can be utilised to obtain
consecutive or perpendicular connectivity between two
assembly lines.

The consecutive connectivity can be benefited when
two assembly lines are located consecutively. This
connectivity type can appear between an upstream line
(supplier) and its downstream line (customer). If the
output of an upstream line is the main input (part) of a
downstream line, then assembly line managers may
desire to locate these lines consecutively and close to
each other. This means that the output of the upstream
line is processed throughout most of the tasks on the
downstream line. In this case, we will have the chance
to obtain a consecutive connectivity between these
lines. Figure 2 shows two simultaneously balanced
consecutive assembly lines.
As shown in Figure 2, two consecutive assembly
lines are connected with a common workstation that
includes tasks 4 and 5 of line 1, and task 1 of line 2.
A consecutive connection must be constructed at the
end of the upstream line and the beginning of the
downstream line.
The outputs of an upstream line may not always be
the main part for a downstream line. In other words,
the outputs of the upstream line may not be processed
throughout most of the tasks on the downstream line.
The outputs may be components that are attached to
the main parts processed on the downstream line. This
attachment can be performed at a stage (task) of
assembly process. In this case, assembly line managers
probably desire to locate the upstream line to the
nearest point of use so as to minimise material
handling from the upstream line to the downstream

line. If such a location exists in a production system,
two lines can be connected to each other with
perpendicular connectivity. Figure 3 shows two assembly lines that have a perpendicular connectivity with a
common workstation.
Figure 3 shows that two lines are connected with a
common workstation which includes task 3 of line 1
and task 5 of line 2. The outputs of the upstream line
are required at the point of task 5 of the downstream
line. That is, task 5 cannot be completed unless the
outputs of line 1 feed it.
Based on the locations, an assembly line of a
production system can be connected to another


International Journal of Computer Integrated Manufacturing
assembly line with parallel, consecutive or perpendicular connectivity. Two assembly lines can be connected to each other with only one of these
connectivity types at the same time. However, an
assembly line can be connected to several assembly
lines with different connectivity types at the same time.
Such connectivity cases can be called a mixed
connectivity. Figure 4 shows a mixed connectivity of
four assembly lines.
Figure 4 shows that line 1 and line 2 are connected
to each other with a parallel connectivity; line 2 and
line 3 are connected to each other with a perpendicular
connectivity; line 2 and line 4 are connected to each
other with a consecutive connectivity. As shown in
Figure 4, line 2 is connected to three lines with all types
of connectivity at the same time. In addition, line 1 is
not allowed to connect to line 3 and line 4.

Balancing several assembly lines simultaneously
using their connectivity opportunities will provide
flexibility to minimise the total idle times of the lines
and total number of workstations required in the
production system. However, there is an important
problem when connecting multiple assembly lines with
common workstations. If the cycle times of assembly
lines vary, the cycle time for a common workstation is
the minimum of the cycle times of multiple assembly
lines that it spans (Miltenburg 1998). For example,
consider a common workstation that spans line a,
where the cycle time is 5 min and that spans line b,
where the cycle time is 10 min. If the workload of this
workstation exceeds 5 min, the operator will be unable
to complete tasks on line a in its cycle time.
3.

Figure 3.

A perpendicular connectivity.

Figure 4.

A mixed connectivity.

405

Mathematical formulation

The most important management problem in assembly
lines is the Assembly Line Balancing (ALB). The
simplest version of ALB problems is the SALB
problem mentioned above. ALB is the problem of
assigning assembly tasks to successive workstations by
satisfying some constraints and optimising a performance measure. This performance measure is usually
the minimisation of the number of workstations
utilised in the assembly line. For a given cycle time,
minimising the sum of station idle times is equal to
minimising the number of opened stations. SALB
problems under this objective are called Type-1.
Conversely, if the number of stations is given, then
minimising the cycle time guarantees minimum idle
times, which is known as Type-2. If both number of
stations and the cycle time can be altered, the line
efficiency E is used to determine the quality of a


406

H. Go¨kc¸en et al.

balance. The corresponding SALB problem was hence
labeled Type-E (Boysen et al. 2007). It is usually
assumed in ALB literature that each workstation
requires one worker. Hence, if the number of workstations utilised in the line is minimised then the
number of workers required is also minimised and
productivity is maximised. There are three main
constraints of ALB problems. The first set of
constraints is called assignment constraints that ensure

each task is assigned to at least and at most one
workstation. Cycle time constraints are the second set
of constraints that guarantee the workload of a
workstation will not exceed the cycle time. Cycle time
is the time interval between two successive completed
products and represents the output rate of the line. In
other words, an assembly line produces one product
for each cycle time unit. Therefore, the workloads
of workstations should not exceed cycle time. The
precedence relationships among tasks are satisfied by
precedence constraints. The precedence relationships
among assembly tasks are illustrated using precedence
diagrams. Figure 5 shows an example precedence
diagram with six tasks.
In this section, we propose a binary formulation for
MSLB. The proposed formulation attempts to balance
an entire production system which consists of

Figure 5.

An example precedence diagram.

Figure 6.

An integrated precedence diagram.

combined assembly lines. The proposed formulation
is aiming at integrated balancing of multiple assembly
lines to attain Shojinka in a multiple straight line
facility using the connectivity types mentioned in

Section 2. Therefore, we need an integrated precedence
diagram that represents all precedence diagrams of
multiple assembly lines. We suggest that the integrated
precedence diagram of a production system can be
obtained by merging individual precedence diagrams
into a huge diagram. Multiple precedence diagrams
can be integrated based on the connectivity opportunities between multiple assembly lines. For example, if
a consecutive connectivity exists between two assembly
lines, the precedence diagram of the downstream
assembly line can be appended at the end of the
upstream assembly line. Thus, these assembly lines can
be connected with a common workstation which
includes at least the last task of upstream line and
the first task of downstream line. Based on the same
connectivity opportunities given in Section 2, Figure 6
shows the integrated precedence diagram of multiple
assembly lines shown in Figure 4.
Although the production system has a SALB
problem for each line, these SALB problems are now
transformed into a single MSLB problem using the
integrated precedence diagram. It should be noted in
Figure 6 that task numbers are sequentially modified in
the integrated diagram to prevent any inconvenience.
Since line 1 and line 2 are located in parallel and they
can be combined with parallel connectivity opportunities, no precedence relationships between any tasks
of these lines are established in the integrated diagram.
If a supplier–customer relationship exists between an
upstream and a downstream assembly line, a

407

International Journal of Computer Integrated Manufacturing
precedence relationship between two assembly lines
should be established. There is no supplier–customer
relationship between line 1 and line 2. Once the
integrated precedence diagram is established and
connectivity options between assembly lines are identified, the following formulation can be used for
simultaneous balancing of multiple assembly lines:
Indices:
i,r,s,k,l
j
h,g,a,b,p,q

We assume in our formulation that completion
times of tasks are deterministic and the longest task
time is not greater than cycle time. The mathematical
formulation is presented below:
K
max
X

Minimise

zj

ð1Þ

8i 2 Ih

ð2Þ

j¼1

Subject to:
: task
: workstation
: assembly line

Parameters and sets:
Ih
: set of tasks on line h
J
: set of workstations; j ¼ 1,. . .,Kmax
A
: set of assembly lines
nh
: number of tasks on line h
H
: total number of assembly lines in the
production system
N
: total number of tasks
P in the
production system; N ¼ 8h2A nh
Kmax
: maximum number of workstations
Ch
: cycle time of line h
ti

: completion time of task i
M
: a big number
S
: set of all precedence relationships in
the production system
(r,s)2 S : a precedence relationship; task r is an
immediate predecessor of task s
R
: set of disconnection relationships
(p,q)2R : a disconnection relationship; line p
cannot be connected with line q
F
: set of perpendicular connectivity
relationships
(a,b)2F : a pair of lines; upstream line a can be
connected to downstream line b with
perpendicular connectivity
k
: the last task in the precedence diagram
of upstream line a
l
: the task of downstream line b to which
the outputs of upstream line a are input
Variables:
xhij
: 1, if task i of line h is assigned to
workstation j; 0, otherwise
Uhj
: 1, if workstation j is utilised on line h; 0,

otherwise
zj
: 1, if workstation j is utilised; 0, otherwise
W(a,b)j : 1, if workstation j includes tasks from
both line a and line b. That is, it is a
common workstation; 0, otherwise
V(a,b)j : 1, if workstation j includes tasks from
only line a or line b. That is, it is not a
common workstation; 0, otherwise

K
max
X

xhij ¼ 1

8h 2 A

j¼1
K
max
X

ðKmax À j þ 1Þðxhrj À xhsj Þ ! 0

j¼1

ð3Þ

8ðr; sÞ 2 S

N
X

M À ðM À Cg ÞUgj

ti xhij

i¼1

8j 2 J
X

8g 2 A

xhij À nh Uhj

08j 2 J

8h 2 A

i2Ih

X

ð5Þ

xhij À nh Uhj ! 1 À nh
ð6Þ

i2Ih

8j 2 J

8h 2 A

Upj þ Uqj

8j 2 J8ðp; qÞ 2 R

1

Uaj þ Ubj À 2Wða;bÞj À Vða;bÞj ¼ 0
8j 2 J
K
max
X

ð4Þ

8ða; bÞ 2 F

Wða;bÞj

8ða; bÞ 2 F

1

ð7Þ

ð8Þ

ð9Þ

j¼1

xakj þ xblj À 2Wða;bÞj ! 0
8j 2 J
H
X

8ða; bÞ 2 F

Uhj À Hzj

0

8j 2 J

ð10Þ

ð11Þ

h¼1

The objective of the proposed formulation is to
minimise the number of workstations utilised in the
entire production system. Equation (2) ensures that
each task is assigned to at least and at most one
workstation. Equation (3) ensures that task s cannot be
assigned until its predecessor task r is assigned to an

earlier or the same workstation that task s is assigned.


408

H. Go¨kc¸en et al.

Equation (4) is the cycle time constraints of the model.
This set of constraints ensures that the work content of
a workstation does not exceed cycle time. These
constraints are designed for the situation that each
line runs at different cycle times. Therefore, these
constraints also ensure that the cycle time for a
common workstation is the minimum of the cycle
times of multiple assembly lines that it spans.
Equations (5) and (6) determine whether workstation
j is utilised on line h or not. If workstation j is utilised
on line h then Uhk will be 1, otherwise it will be 0.
Connecting two assembly lines with any connectivity
type may not always be possible. Therefore, Equation
(7) is used to disconnect two assembly lines that cannot
be connected with one or more common workstations.
Equations (8), (9) and (10) are added to the model to
establish perpendicular connections. For a given (a, b)
perpendicular connectivity relationship, Equation (8)
determines whether any common workstation j between lines a and b is utilised or not. Equation (9)
ensures that at most one common workstation can be
utilised between lines a and b. If a common workstation between lines a and b is utilised, Equation (10)
guarantees this workstation contains the last task (k)
of the upstream line a and the task (l) of the

downstream line b to which the outputs of upstream
line a are the input. Equation (11) determines whether
workstation j is utilised or not.

Suppose assembly line managers conveyed that
there are several connectivity opportunities between
four assembly lines. These opportunities are given in
Table 2.
The set of disconnection relationships can be easily
obtained using Table 2. Using these connectivity
opportunities, the integrated precedence diagram is
constructed and presented in Figure 7.
At the first stage, each assembly line is balanced
independently supposing no connectivity between
assembly lines is allowed. The results for independent
line balances will show us the effects of common
workstations on total number of workstations utilised
in the production facility. The maximum number of
workstations (Kmax) is selected as 6 and the problems
are solved using LINGO (Schrage 2002) on an Intel
Core 2, 2.00 GHz, 1022 MB Ram computer. Independent line balances are given in Table 3.
Table 3 shows that a total of seven workstations
are required for the optimal solution. For workstation
1, 2/5 ¼ 40% of total available workforce is unproductive. The ratios of unproductive times for other
workstations are 40%, 20%, 50%, 30%, 40% and
25% respectively with an average of 35%. This can be
considered as an important waste of workforce
resources. According to the assignments given in

Table 2.

lines.

4.

Connectivity opportunities between assembly

Illustrative examples

The proposed model is illustrated using a simple
multiple straight assembly line facility. The example
facility consists of four assembly lines with a total of
15 tasks. Task completion times and cycle times of
assembly lines are given in Table 1.
Table 1.

2

3

4

–

None
–

Parallel
None
–

None
Perpendicular
Consecutive
–

Task data and cycle times of assembly lines.
Tasks (i)

Task Times (ti)

Cycle Times (Ch)

1

1
2
3

2
1
3

5

2

4
5
6

3
5
4

10

3

7
8
9
10

3
2
4
3

15

4

11
12
13
14
15

4
8

2
3
10

20

Lines (h)

1
2
3
4

1

Figure 7.
example.

The integrated precedence diagram of the


409

International Journal of Computer Integrated Manufacturing
Table 3.

Independent line balances.

Workstation/Line
1

2
3
4
5
6
7

1

2

3

4

Workloads

Idle Times

11,12
13,14,15

3
3
12
5
7
12
15

2
2
3
5
3
8
5

1,2
3
7,8,9,10
5
4,6

Table 3, the layout of the facility can be illustrated as
shown in Figure 8.
The example facility is then balanced considering
the connectivity opportunities given in Table 2. Since
the cycle times of assembly lines vary, the cycle time for
a common workstation will be the minimum of the
cycle times of multiple assembly lines that it spans. In
addition to the precedence relationships for each
individual precedence diagram, (6, 14) and (10, 11)
precedence relationships are added to S for reflecting
the integration of precedence diagrams. The results are
given in Table 4.
Table 4 shows that total number of workstations is
reduced to six when assembly lines are connected to
each other with common workstations. Workstations
1, 3 and 5 are common for two different assembly lines.

Workstation 1 is common for lines 1 and 3, workstation 3 is common for lines 3 and 4, and workstation
5 is common for lines 2 and 4. According to the
assignments given in Table 4, the layout of the facility
can be illustrated in Figure 9.
The cycle times for these common workstations are
the minimum of the cycle times that they span. For
example, the cycle time for line 2 is 10 min and the
cycle time for line 4 is 20 min. However, the cycle time
for workstation 5 that is common for these lines is
10 min. Owing to the timing difficulties and operator/
machine interference, the multi-line (common) workstations that were utilised on these lines are more
difficult to operate when the cycle times of the lines are
varied (Miltenburg 1998). Consider workstation 5 in
Figure 10 as an example. Task 6 must be completed
once every 10 min, while task 14 must be completed
once every 20 min. Therefore, a task sequence that the
operator will follow must be generated to operate this
workstation synchronously. Three Gantt charts for
task sequences of common workstations 1, 3, and 5 are
generated and given in Figure 10.
The Gantt charts show when the operator must
start and finish the tasks and when the operator is idle.
The task sequences in Figure 10 are generated for a
limited duration. This duration is the least common
multiple of the cycle times of associated assembly lines.
For instance, the task sequence of workstation 3 is

Figure 8.

Table 4.

Facility layout for independent line balances.

Line balances with connectivity.

Workstation/Line
1
2
3
4
5
6

Figure 9.

1

2

3

4

1
–
–
2,3
–
–

–
4,5
–
–
6
–

7
–
8,9,10
–
–
–

–
–
11,13
–
14
12,15

Facility layout with common workstations.

generated for 60 min because the least common
multiple of 15 and 20 is 60. After 60 min, the operator
will follow the same task sequence. Therefore, the
magnitude of workloads and idle times for common
workstations must be calculated by considering least

410

Figure 10.

H. Go¨kc¸en et al.

Task sequences for common workstations.

common multiple of cycle times. Figure 10 shows that
6 min of 15 min is idle for workstation 1, 6 min of
60 min is idle for workstation 3 and 9 min of 20 min is
idle for workstation 5. This means that 40%, 10% and
45% of available time of workstations are unproductive respectively. The workloads of regular (single-line)
workstations 2, 4 and 6 are 8, 4 and 18 min respectively
with 20%, 20% and 10% unproductive idle time
ratios. These values mean that the average rate of
unproductive times for the example multiple assembly
line facility is 24.16%. That is, unproductive times for
workstations are reduced significantly compared with
independent line balances.

5. Conclusions
The main objective of the TPS is to increase
productivity by eliminating or reducing all kinds of
wastes in a production system. Shojinka is an
important TPS technique that aims to eliminate or
reduce operators’ idle times. If the cycle time is
sufficiently large, an operator can perform operations
on two or more different production lines at the same
time. The problem of assigning operations to operators

is important while implementing Shojinka. In this
paper, we proposed that if practical, Shojinka can be
implemented in a multiple straight assembly line


International Journal of Computer Integrated Manufacturing
facility. The problem of assigning assembly tasks of
multiple assembly lines to minimum number of
workstations is called MSLB. The MSLB problem is
built on the concept that it should be possible for a
company to obtain the advantages of Shojinka even if
the company has not adopted the U-shaped line
layout. Three connectivity types to combine multiple
straight assembly lines are investigated and explained.
A binary formulation for MSLB is developed based
on the connectivity types investigated. The proposed
formulation assigns some tasks to common workstations which include tasks on two or more
neighbour assembly lines. The utilisation of common
workstations minimises total idle times of assembly
lines and total number of workstations required in the
facility. As with other types of ALB problems, the
MSLB problem is also NP-Hard. Therefore development of effective heuristics to solve an MSLB
problem is an important topic for future researches.
In addition, the proposed binary formulation is
expected to help researchers to develop formulations
for combining multiple U-lines.
References
Aase, G.R., Olson, J.R., and Schniederjans, M.J., 2004. Ushaped assembly line layouts and their impact on labor
productivity: An experimental study. European Journal of
Operational Research, 156 (3), 698–711.

Baybars, I., 1986. A survey of exact algorithms for the simple
line balancing problem. Management Science, 32 (8),
909–932.
Becker, C. and Scholl, A., 2006. A survey on problems
and methods in generalised assembly line balancing.
European Journal of Operational Research, 168 (3), 694–
715.
Boysen, N., Fliedner, M., and Scholl, A., 2007. A classification of assembly line balancing problems. European
Journal of Operational Research, 183 (2), 674–693.

411

Chiang, W.C., Kouvelis, P., and Urban, T.L., 2007. Line
balancing in a just-in-time production environment:
balancing multiple U-lines. IEEE Transactions, 39 (4),
347–359.
Erel, E. and Sarin, S.C., 1998. A survey of the assembly line
balancing procedures. Production Planning and Control, 9
(5), 414–434.
Erel, E., Sabuncuo
glu, I., and Aksu, B.A., 2001. Balancing of
U-type assembly systems using simulated annealing.
International Journal of Production Research, 39 (13),
3003–3015.
Ghosh, S. and Gagnon, J., 1989. A comprehensive literature
review and analysis of the design, balancing and
scheduling of assembly systems. International Journal of
Production Research, 27 (4), 637–670.
Go¨kc¸en, H., A
gpak, K., and Benzer, R., 2006. Balancing of

parallel assembly lines. International Journal of Production Economics, 103 (2), 600–609.
Go¨kc¸en, H. and A
gpak, K., 2006. A goal programming
approach to simple U-line balancing problem. European
Journal of Operational Research, 171 (2), 577–585.
Miltenburg, J. and Wijngaard, J., 1994. The U-line balancing
problem. Management Science, 40 (10), 1378–1388.
Miltenburg, J., 1998. Balancing U-lines in a multiple U-line
facility. European Journal of Operational Research, 109
(1), 1–23.
Monden, Y., 1993. Toyota production system. Norcross, GA:
Engineering and Management Press.
Salveson, M.E., 1955. The assembly line balancing problem.
Journal of Industrial Engineering, 6 (3), 18–25.
Scholl, A. and Klein, R., 1999. Ulino: Optimally balancing
U-shaped JIT assembly lines. International Journal of
Production Research, 37 (4), 721–736.
Schrage, L., 2002. LINGO release 8.0. LINGO System Inc.
Sennott, L.I., Oyen, M.P.V., and Iravani, S.M.R., 2006.
Optimal dynamic assignment of a flexible worker on an
open production line with specialists. European Journal
of Operational Research, 170 (2), 541–566.
Sparling, D., 1998. Balancing Just-In-Time production units:
the N U-line balancing problem. Information Systems and
Operational Research, 36 (4), 215–237.
Urban, T.L., 1998. Optimal balancing of U-shaped assembly
lines. Management Science, 44 (5), 738–741.


International Journal of Computer Integrated Manufacturing

Vol. 23, No. 5, May 2010, 412–424

Job shop scheduling by pheromone approach in a dynamic environment
P. Renna*
DIFA, University of Basilicata, Via dell’Ateneo Lucano, 10, Potenza, Italy
(Received 19 July 2009; final version received 5 January 2010)
Job shop scheduling problem is a NP-hard problem; therefore the objective is to create a schedule that satisfies all
the constraints while taking as little overall time as possible. The paper concerns the job shop scheduling problem in
cellular manufacturing systems; the schedule is created by a pheromone-based approach. The proposed approach is
carried out by a Multi-agent Architecture and it is compared with a coordination approach proposed in literature
used as a benchmark.
A simulation environment developed in ARENA1 package was used to implement the approaches and evaluate
the performance measures. The performance measures investigated are: throughput time, throughput, Work In
Process, machines average utilisation and tardiness. Several scenarios are considered: from static to very dynamic
conditions for internal and external exceptions of the manufacturing system. The simulation results highlighted that
the performance of the proposed approach are comparable with the benchmark when the customer demand has a
high fluctuation and the manufacturing system is less dynamic.
Keywords: dynamic scheduling; ant colony intelligence; pheromone; multi-agent systems; discrete event simulation

1.

Introduction

The environment within which manufacturing systems
operate is characterised by more rapid change than
ever before. The unforeseen disturbances occur frequently and the manufacturing system has to be able to
react to these disturbances. The disturbances can be:
machine breakdowns, demand variability and delay in
processing time. The above reasons drive the manufacturing systems to adopt dynamic scheduling
approaches.

Dynamic scheduling has been defined under four
categories (Mehta and Uzsoy 1999, Vieira et al. 2000a,
2003;, Aytug et al. 2005, Leus and Herroelen 2005):
(a) completely reactive scheduling; in this case no
schedule in advance is created and the job
schedule is obtained in a real-time fashion.
(b) predictive-reactive scheduling; a schedule is
created in advance, but a rescheduling is
considering to respond to exceptions in real
time.
(c) robust predictive-reactive scheduling; a schedule
is created in advance and rescheduling activity
is activated when the effect on the performance
measures of the exceptions is significantly.
(d) robust pro-active scheduling; the schedule is
computed in advance predicting the effect of
exceptions on the manufacturing system.

*Email:
ISSN 0951-192X print/ISSN 1362-3052 online
Ó 2010 Taylor & Francis
DOI: 10.1080/09511921003642170


The dynamic scheduling problem can be solved by
using the following techniques:
(a)
(b)
(c)
(d)

mathematical programming approach;
dispatching rules approach;
heuristic approach;
artificial intelligence approach.

Moreover, most of the scheduling systems developed
in industrial environments are centralised and hierarchical, but these approaches present several drawbacks. The most important drawbacks are the
following:
(a) a central computer; it constitutes a bottleneck
with a limit of capacity and a failure of it leads
to bring down the entire manufacturing system;
(b) it is more difficult to extend and modify the
configuration of the scheduling system;
(c) a slow response to disturbance because the
information has to flow to the high level and
then a reaction is planned.
Therefore, a centralised and hierarchical scheduling is
inefficient in a very dynamic environment where the
exceptions are more frequent.
For the above reasons, a decentralised control
method is more efficient in a dynamic environment in


International Journal of Computer Integrated Manufacturing
particular, Multi-Agent System (MAS )methodology is
more suitable for the implementation of decentralised
system. MAS can be defined as a network of problem
solvers that work together to solve problems that are
beyond their individual capabilities (O’Hare and

Jennings 1996).
MAS approaches are more suitable to develop agile
and robust distributed control, but its performance
depends more on the coordination mechanism. The
coordination mechanisms proposed in literature are
the Contract net protocol (Smith 1980), market-based,
auction-based (Siwamogsatham and Saygin 2004) and
game theory. These approaches have same limitations
as: communication overhead, constantly exchange of
information and therefore a minor reactivity of the
agents.
Recently, many authors have developed several
approaches inspired by the behaviour of social insects
such as ants, bees, termites and wasps to propose an
alternative method for coordination in complex
systems. The most promising approach is based on
Ant Colony; Ant Colony coordination refers to the
cooperative ant foraging behaviour. Ant colonies can
always find shorter paths from a nest to a food source
by pheromone trail laying and following. Dorigo et al.
(1991) first introduced Ant Colony Optimisation
(ACO) for solving the Travelling Salesman Problem,
which is based on ant foraging. In manufacturing
scheduling problem, the approach can be inspired to
pheromone-based rule of Ant Colony to coordinate the
MAS architecture.
The aim of this paper is to investigate the
performance of a pheromone approach for cellular
manufacturing systems in a very dynamic environment. The pheromone approaches proposed are two:
one based on moving average and the other one on

exponential moving average.
Two different types of disturbances are considered.
(1) Internal exceptions; these exceptions are caused
by the resources of the manufacturing system,
in particular, they are: machine breakdowns
and efficiency of the manufacturing machine.
(2) External exceptions; these exceptions are
caused by external changes of the manufacturing system, specifically they are: changes in
volume and mix demand products.
Moreover, it is considered the effect of dispatching rule
on the performance of the manufacturing system. The
performance ares compared with an approach proposed in literature based on classical coordination
mechanism in MAS.
The structure of the paper is as follows: in
Section 2 an overview of literature is presented. The

413

manufacturing context is described in Section 3. In
Section 4 the proposed coordination approaches based
on pheromone are explained. The simulation environment and design of experiment conducted are given in
Section 5, while in Section 6 the simulation results are
presented. Finally, conclusions and a future research
path are presented in Section 7.
2.

Literature review

Recently, a few researchers have developed approaches
based on Ant Colony inspired scheduling in shop floor

control. Most of the applications proposed concern the
shop floor routing and permutation flow-shop sequencing problem.
Peeters et al. (2001) has developed a concept
control system based on coordination mechanism of
insect colonies, in particular, a pheromone-based
control scheme is introduced. Basic principles of the
pheromone concept, the control system architecture
and a layered approach for decision-making have
been discussed. Test beds of industrial scale have
been used to demonstrate properties and benefits
of this approach. The main advantages that emerged
for the pheromone approaches are: a simple coordination mechanism; the automatic guidance to the
optimised solution; the capability to handle dynamic
situations.
Yu and Ram (2006) proposed a multi-agent
approach designed for dynamic job shops with routing
flexibility and sequence-dependent setup. A bio-inspired strategy based on division of labour in insect
societies is presented for coordination among agents.
The strategy is accomplished using a computational
model, which is composed of response threshold,
response intention, and machine-centred reinforcement
learning. The bio-inspired scheduling is compared with
an agent-based approach and a dispatching rule-based
approach. The experiments were performed using
simulation and statistical analysis. Results show that
the proposed bio-inspired scheduling model performs
better than the other two methods on all eight common
scheduling metrics.
Xiang and Lee (2008) presented research concerning dynamic scheduling by agent coordination that is
inspired on Ant Colony in MAS. The proposed

approach is tested in a shop floor model considers
multiple job types and parallel multi-purpose machines
with dependent setup times. Moreover, the machine
break down is included as a disturbance of the
manufacturing system. In this research, the Ant
Colony is used to find an appropriate machine agent
for processing and helps the machine agent to
determine the next job to be processed in the current
queue.


414

P. Renna

Scholz-Reiter et al. (2008) proposed a pheromonebased autonomous control method to a mix model of a
shop floor and tested it in different dynamic demand
situations. They compared the pheromone approach
with a method based on queue length estimator. The
only performance is the throughput time with a
sinusoidal arrival rate of the parts.
Zhou et al. (2009) proposed an algorithm based on
Ant Colony Optimisation in a shop-floor scenario with
three levels of machine utilisation, three different
processing time distributions, and three different
performance measures for intermediate scheduling
problems. The performances measured are: mean
flow time, mean tardiness, total throughput on
different experimental environments compared with
those from dispatching rules including first-in-first-out,

shortest processing time, and minimum slack time. The
experimental results show that ACO outperforms
other approaches when the machine utilisation or the
variation of processing times is not high. The
procedure proposed is a centralised approach.
Renna (2009) developed two pheromone approaches for the job shop scheduling problem. One
is based on the past information of a part
(throughput time of the manufacturing cell); and
the other on the queue of the manufacturing cell.
The proposed approaches are tested in dynamic
environment; the simulation results show how the
approach based on the queue of the manufacturing
cell performs better when the environment conditions
are very dynamic.
Based on the above literature review, the following
limitations can be drawn:
(a) a benchmark based on MAS architecture with
complex coordination mechanism was not used
to evaluate the advantages and disadvantages
of Ant Colony coordination mechanisms;
(b) the proposed approaches in literature are tested
in manufacturing systems in which some
exceptions occur and the analysis on the
rapidity of alterations was not investigated.
(c) each model proposed is tested on few performance measures (for example only throughput
performance).
The main findings of the research proposed in this
paper can be summarised as follows:
(a) the pheromone approach is deeply investigated
by an adequate number of performance

measures;
(b) the simulations have been conducted in a
several dynamicity levels introducing both
internal and external disturbances;

(c) the pheromone approaches have been compared with a complex MAS approach in order
to find out in which environment conditions the
proposed approach can be competitive.
3.

Manufacturing system context

The manufacturing system consists of a given number
of cells; each cell is able to perform any kind of
manufacturing operation so that the resulting manufacturing system is a pure general-purpose one. In
such a system, the scheduling decision consists in
deciding in what manufacturing cell the part will
perform the next operation, therefore it is a pure
dispatching problem. The manufacturing system has
the following characteristics:
(1) Several part types have to be manufactured by
the manufacturing system. Each part type has a
predefined number of operations performed by
the manufacturing cells. At each part is
assigned a due date.
(2) Orders for production of different parts arrive
randomly with an inter-arrival that is exponential distribution;
(3) Each machine performs the manufacturing
operation with an efficiency, which sets the
speed of the operation.

(4) The queues are managed by the First In First
Out policy in order to investigate only the
pheromone approaches policy.
(5) Each machine can breakdown randomly with
an exponential distribution.
In this research, the transportation time of the material
handling devices are included in the processing time,
and the handling resources are always available.
The coordination approach based on Ant Colony is
inspired by the behaviour of foraging ants that leave a
pheromone trail on their way to the food. In the real
world, ants (initially) wander randomly, and upon
finding food return to their colony while laying down
pheromone trails. If other ants find such a path, they
are likely not to keep traveling at random, but instead
follow the trail, returning and reinforcing it if they
eventually find food. Over time, however, the pheromone trail starts to evaporate, thus reducing its
attractive strength.
Ant Colony is a natural example of a highly
distributed system, therefore, it can be formalised by
MAS. The coordination mechanism based on Ant
Colony has the following advantages.
(1) The simplicity of the coordination mechanism.
In fact, the ants do not communicate directly