iBusiness, 2010, 2, 201-209
doi:10.4236/ib.2010.23025 Published Online September 2010 (http://www.SciRP.org/journal/ib)
Copyright © 2010 SciRes. iB
The Holonic Production System: A Multi Agent
Simulation Approach*
Gandolfo Dominici1, Pierluigi Argoneto2, Paolo Renna2, Luigi Cuccia3
1S.E.A.F., University of Palermo, Faculty of Economics, Palermo, Italy; 2D.I.F.A., University of Basilicata, Faculty of Engineering,
Basilicata, Italy; 3D.T.M.P.I.G., University of Palermo, Faculty of Engineering, Palermo, Italy.
Email: gandolfodominici@unipa.it, paolo_renna@yahoo.it, {pieroargoneto, luigi.cuccia}@gmail.com
Received May 19th, 2010; revised June 3rd, 2010; accepted July 29th, 2010.
Today’s turbulent markets are facing unpredictable and sudden variations in demand. In this context, the Holonic Pro-
duction System (HPS) seems to be able to overcome the operational and economic problems of traditional production
systems. The HPS’ ability to adapt and react to business environment changes, whilst maintaining systemic synergies
and coordination, leverage on its network organizational structure, assuring both flexibility and profitability. In this
paper we study HPS experimentally, modeling holon-firms as agents. In our simulation, holon-firms interact both with
each other and with the external environment without predetermined hierarchies and following their own aims and in-
ternal decision rules with a negotiation-based control system. The Multi Agent System Approach we propose aims to
evaluate and test the performance of the HPS to adjust to changes in market demand by simulating variations in
holon-firms’ capacity and reconfiguration costs in real time in a distributed enterprise network. Hence we demonstrate
that, through a collaborative negotiation approach, the HPS results in a better adaptability and improved network re-
Keywords: Holonic Production System (HPS), Multi Agent System (MAS), Distributed Enterprise Network
1. Introduction and Theoretical Framework:
Why the HPS?
Mass production proved its effectiveness in stable envi-
ronments with continuous growth trends, until the end of
the 1980’s. Before that time, the hierarchical pattern on
which mass production is founded, assumed the steadi-
ness of social, economic and technological factors. Since
the beginning of the 1990’s, this pattern begun to show
its weaknesses due to the increasing instability and the
growing systemic complexity of business environments.
The spread of Internet made it possible for firms to use a
low cost, worldwide extended, informative infrastructure
which brought profound changes in the market. Informa-
tion and Communication Technology (ICT) allows to
develop production networks between firms which can
be delocalized and dispersed in space. These changes
caused the shift from “mass production” to “mass-cus-
tomisation”. In turbulent markets, the reduced flexibility
of mass production due to hierarchical control systems
leads almost inevitably to criticalities. The problems
concerned with unpredictable variations of demand rate
unnecessary increases in capacity and inventory
level, which reduce efficiency;
decline of service level, resulting in a negative im-
pact on customer satisfaction.
In order to fulfil the new needs for agility, it becomes
unavoidable for firms to develop an extremely flexible
production structure able to [1,2]:
1) react to the market environment’s turbulences;
2) survive to the changes of production system through
the adoption of new technologies;
3) adapt to the uncertainties of production systems in
such environments.
The literature on this topic shows several trends which
production and supply chain systems have to adapt to
1) the paradigm shift from mass production to semi-
personalized production;
2) the opening to collaboration with other agents so as
to speed up production innovation and processes;
3) the decisive role of effective and efficient cooper-
*The manuscript was contributed by four authors, G. Dominici con-
tributed Section 1; P. Argoneto contributed Sections 2, 4 and 6; P.
Renna contributed Sections 2, 5 and 6; L. Cuccia contributed Sections 6
and 7.
The Holonic Production System: A Multi Agent Simulation Approach
Copyright © 2010 SciRes. iB
ation within networks;
4) the awareness of the problems related to the imple-
mentation of centralized control systems consider-
ing the differences of information, experiences, ac-
tivities and objectives among the entities partici-
pating to the same network.
These changes call for new organizational structures.
Traditional hierarchical systems show several shortfalls
in these new business environments:
1) they strongly limit the reconfiguration, the reliabil-
ity and the growth capacities of the organization [4];
2) their complexity grows together with the size of the
organization [5];
3) communication among the elements of the system is
rigorously determined ex ante and vertically limited
4) the structure’s modules cannot take initiatives, thus
reducing the system’s readiness to react therefore
resulting not agile in turbulent environments [7];
5) the structure is expensive to build and to preserve.
Heterachical systems do not have the limitations of hi-
erarchical systems, as they are able to maintain flexibility
and adaptability to external stimuli. In heterachical sys-
tems, hierarchy is excluded and control is managed by
the single “agents” of the system. Agents interrelate with
their environment and with other agents according to
their own characteristics and scopes. In this system de-
void of prearranged hierarchies, control is based on ne-
gotiation [8].
In the field of artificial intelligence, the term agent is
used to identify the intelligent elements of a system, who
act in the environment as entities able to be conscious of
situations and who pursue an aim; such agents must en-
compass the following attributes [8,9]:
autonomy, they act without the help or guide of any
superior entity;
social ability, they interact with other agents;
reactivity, they perceive their environment and re-
spond rapidly to changes;
pro-activity, they are able to have initiative and spe-
cific behaviors for a specific scope.
Though they are agile, heterachical systems are not
capable of operating according to predefined plans, for
this reason their behavior is hardly predictable, thus in-
creasing unpredictability in systemic dynamics. Heterar-
chical structures work well in simple, non complex and
homogeneous environments with abundance of resources
[7], while in complex environments with shortage of re-
sources they can cause instability and waste. Hence it is
necessary to create a system which is able to assure both
performance and reactivity at the same time.
Theories on living organisms and social organizations
have been used as interpretive lenses to analyse business
systems and firms’ networks. In this view organizations
are represented as living systems. The holonic paradigm
emerges within this view, stemming from the holistic1
approach and the viable system approach [10]. The holo-
nic paradigm originates from the thought of Arthur
Koestler [11] that highlights how complex systems de-
rive from the union of stable and autonomous
sub-systems, which are able to survive turbulences and,
at the same time, cooperate to shape a more complex
system. Koestler underlines that the analysis of both the
biological and the physical world shows that it is neces-
sary to take into account the links between the whole and
the part of the entities we observe. According to Koestler
in order to understand complex systems, it is not enough
to study atoms, molecules, cells, individuals or systems
as independent entities, but it is essential to consider such
unities as concurrently parts of a larger whole; in other
words, we have to consider the holon. The term holon is
a blend of the ancient Greek “λος with the meaning of
“whole” and the suffix “ν” meaning entity or part; thus
the whole is set up of parts which, unlike atoms, are also
entities [2]. The holon is, in fact, a whole which includes,
simultaneously, the elements or the sub-parts which stru-
cture it and give it a functional meaning. Holons act as
intelligent, autonomous and cooperative entities which
work together inside temporary hierarchies called holar-
chies. A holarchy is a hierarchy of self-regulating holons
working, in harmony with their environment, as autono-
mous wholes which are hierarchically superior to their
own parts and, in unison, are parts dependent on the con-
trol of superior levels. There are three pillars of holonic
system [12]:
1) a shared-value system which allows the spontane-
ous and continuous interaction among groups of
people who are geographically dispersed and are not
linked by legal or ownership ties, consenting to ac-
cess the economies of cooperation and increasing
the stability of the system. Examples of shared value
systems are some of the elements of lean production,
that are often embedded in the company’s vision,
such as the principle of continuous improvement
2) a distributed network information system which is a
neural sub-system [13] supporting real time supply
of information between operating units. Such a sys-
tem consents the pursuit of maximum income by
increasing the capacity to sense and exploit new
business opportunities;
3) an autonomous distributed hierarchy, which is bas-
ed on the ability of each autonomous part to become
leader according to the requirements of specific
1Holistic scientific paradigm focusing on the study of Complex Adap-
tive Systems (C.A.S.).
The Holonic Production System: A Multi Agent Simulation Approach
Copyright © 2010 SciRes. iB
situations caused by the turbulent changes in the en-
vironment. Every entity is able to directly interact
with other entities without mediation. Due to this
property of holonic systems, every holon has poten-
tially the same importance and the same responsi-
bility; the involvement of a holon as operative unit
is based on its knowledge and competencies and is
not a consequence of predefined leadership (Figure
As the reader can see, in the scheme proposed in Fig-
ure 1, the holons are represented as networked agents
which define the production system through coordination
and cooperation [2]. Holons can act as agents of three
different functions: mediator, request and offer functions.
The holon converts its function due to the necessity of
mediation (mediator) or following its state of over (offer)
or under (request) loaded of production capacity.
The mediator function is a negotiation function, which
is distributed among holons. The mediator-negotiation
function has a particular relevance for our simulation
because it manages the distribution of resources among
holons. We consider the mediator-negotiation function as
a primary function and we give it a central role in our
Thanks to computer simulation methodology, it is pos-
sible to test the performance of complex systems scien-
tifically. According to the scheme in Figure 1, we pro-
pose a simulation, based on a multi agent architecture
and a negotiation protocol (which shares the capacity
among the holon-firms), in order to test the proposed
approach comparing it with the “central planner with full
knowledge”, which is here considered as benchmark.
2. Model Formulation
We represent firms as holons (holon-firms). The scenario
concerns I independent and geographically dispersed
holon-firms, each of them able to produce K different
product families, after an appropriate reconfiguration.
Assuming that the total planning horizon T is divided
into N sub-periods t(t = 1N); at the beginning of each
sub-period the holon-firms make their capacity allocation
plan, after having collected backlog and forecast orders
according to the capacity they hold and to the demand
they have to supply. If the holon-firm capacity is not
enough to supply the demand orders, then they can nego-
tiate a portion of capacity with holon-firms whose pro-
duction capacity exceeds their actual demand. This proc-
ess leads to a sharing process in which each holon-firm
tries to maximize its own profit. By doing so, more va-
lue-added collaborations are generated without reducing
the individual benefits of the holon-firm. Specifically,
each holon-firm (f) computes its accomplishment capac-
ity index (ACI) Sf,ACI by measuring the difference be-
tween its own workload (WL) and its capacity (C):
Sf,ACI = WLf – C
f (1)
Subsequently, holon-firms are classified in overloaded
holon-firms (OF), i.e., holons with Sf,ACI > 0, or under-
loaded holon-firms (UF), i.e., holons with Sf,ACI < 0. Af-
terwards, holons belonging to OF or UF compute respec-
tively their required capacity (RCf) or the capacity they
can offer (OCf):
RCf = WLf – C
f (2)
OCf = Cf – WLf (3)
Figure 1. Simplified architecture of a holonic system.
The Holonic Production System: A Multi Agent Simulation Approach
Copyright © 2010 SciRes. iB
As showed in Figure 1, each holon operates by using a
firm agent with a specific function. The mediator func-
tion performs the interaction activities among all the
holons: the mediation aims to share the capacity in order
to maximize the overall profit of the holonic network.
Each holon has the following local knowledge:
Cik: i-th holon-firm’s capacity to produce the k-th
Costik: i-th holon-firm’s total production costs to
produce the k-th product;
Priceik: i-th holon-firm’s unit product price to sell
in k-th holon-firm market ;
WLik: i-th holon-firm’s workload regarding the k-th
Value of OCf or RCf.
2.1. Multi Agent Architecture and Negotiation
We adopt the following multi agent architecture to ana-
lyze the HPS negotiation’s capacities:
each holon-firm belonging to OF is represented by
an OCf Function (OCF) which is in charge for ne-
gotiating the capacity with the holon-firms requiring
it (RCFs);
a mediator function (MF) allows communication
and coordination between capacities of supplier and
applicant holon-firms;
the negotiation process and interaction workflow
between supplier and requester holon-firms is rep-
resented by the UML activity diagram in Figure 2.
The UML diagram is divided in three swim lines, cor-
responding to the requester, supplier and mediator func-
tions. The first activity is performed by the MF that col-
lects RCF capacities requests, sorted by decreasing
amount of capacity. Starting from the higher capacity
request, the MF transmits the proposals to OCF holons-
firms in terms of capacities and prices of each quantity.
The OCF evaluates requester proposals, in terms of quan-
ates t
Wa it
ranking lists
proposal eva
uates co-
ts counte
proposal or accept
comput es
accept or requ
est new counter-proposal
Roun d
coun ter-propos al
Figure 2. Negotiation approach: UML activity diagram.
The Holonic Production System: A Multi Agent Simulation Approach
Copyright © 2010 SciRes. iB
tity and prices, by using the firm’s utility functions.
OCFs can decide: to quit the negotiation, to accept the
proposal or to requests for a new counterproposal when
the current step of negotiation is lower than the estab-
lished maximum one. If the proposal is accepted the
agreement is reached, otherwise the RCF evaluates the
counter-proposal by using its own utility function. There-
fore, the negotiation procedure can be interpreted as a
simple auction based negotiation protocol. It defines the
environmental relations of the autonomous firms in-
volved in the holonic network, but no indication about
the holon-firms decision making behavior (utility func-
tion) is given. This means that the negotiation protocol
can be adapted to different productive functions.
The HPS collaborative network model adopted is bas-
ed on a negotiation approach whilst a centralised alloca-
tion mechanism is used as benchmark.
The negotiation process is characterized by the fol-
lowing constraints:
the negotiation is multi-lateral and involves one to
many holon-firms;
the negotiation is an iterative process with a maxi-
mum number of rounds, rmax, after that an agree-
ment is reached or the negotiation fails;
during each round (r), the OCF can submit a new
counter-proposal (N) to the RCF while, at r = rmax,
it can only accept or reject;
the agreement is reached only if the RCF accepts
the OCF counter proposal at round r < rmax; a con-
tract of production collaboration between the sup-
plier and applicant holon-firms follows the agree-
ment; if there are multiple agreements, the first OCF
that satisfies the RCF sings the agreement.
the holon-firms’ behavior is assumed to be rational
according to its utility function (as defined later);
the RCF and the OCF are mutually unaware of each
others utility function.
The OCF activities can be described as follows:
Wait: the agent is in an initial state of waiting for a
proposal (from RCF);
Evaluates proposal: the OCF evaluates the pro-
posal of the RCF in terms of required capacity and
offered price. At the first round the OCF, commu-
nicates the amount of capacity it is willing to offer
(the minimum value between the one requested by
the RCF and its own unused capacity). Subsequently,
the OCF communicates to the RCF if it accepts or
refuses the proposed price to exchange the promise
amount of capacity. Then the OCF evaluates the
proposal of the RCF on the basis of a threshold
function given by (4):
 
kr kkk k
rice tM
being min( ,)
ij i
RC OC (5)
Expression (4) computed by the OCF is a threshold
level. Starting from the market price value, during the
negotiation the variable r increments and the threshold
level decrease until the value of production costs. In this
case the generated profit is null.
At this point the value of the offer is checked accord-
ing to the following expression:
,,kr kr
valval (6)
If (6) is verified, the jth holon-firm supplies the re-
quested capacity of the ith holon-firm, in this way they
reach an agreement and each of them can update their
available capacity.
Threshold level updates: when the OCF refuses
the price submitted by the RCF, it updates the
threshold level for the next round of negotiation (in-
creases the value of r in expression (4)); if the algo-
rithm reached the last round of negotiation, the OCF
simply quits the negotiation.
Capacity updates: when the negotiation reaches an
agreement, the OCF updates the capacity owned. If
no more capacity resources are available, OCF quits,
otherwise it goes in Wait state.
The activities performed by RCA can be described as
Proposal elaboration: the RCF elaborates a pro-
posal in terms of price and amount of capacity to
acquire, and transmits this information to the MF.
The submitted price is obtained by the following
 
kr kkk k
valprice pricetM
Expression (7) is computed by the RCF and starts
with a price equal to production costs, where the
generated profit is the same obtained when products
are produced by its own holon-firm. During the ne-
gotiation, the price is increased until the market
price. In this case, the generated profit is null.
Wait: the RCF waits for counter-proposal by the
Counter-proposal computation: if the OCF re-
fuses the proposal and the negotiation is still run-
ning, the RCF computes a new counter-proposal
(increasing the value of r in expression (7)). Other-
wise (i.e., it is the last round of negotiation), the
process ends with no agreement.
Capacity updates: if the negotiation reaches an
agreement, the RCF updates its information; if the
acquired capacity is exactly equal to the required
one, it quits, otherwise it computes a new proposal
The Holonic Production System: A Multi Agent Simulation Approach
Copyright © 2010 SciRes. iB
for the residual capacity needed.
The MF coordination activities between OCF and RCF
can be described as follows:
Wait: the MF is in its initial state of waiting for a
proposal (from the RCF).
Computation of ranking list: the MF computes a
ranking list among all the holon-firms that requested
capacity. The way it does it depends on several
variables; in our simulation we consider that the
ranking is done favoring holon-firms with high need
of capacity to allow them to better satisfy custom-
ers’ requests.
Proposal transmition: the MF transmits the pro-
posal computed by RCF to the ranking list of OCF.
Wait: the MF is waiting for the counter-proposal by
all the OCF.
Counter-proposal transmition: the MF transmits
the counter-proposal of the OCF to the RCF.
After all the necessary values have been uploaded, the
holon-firm that does not reach the entire capacity it needs
is inserted in the ranking list again, and the negotiation
starts a new round. To avoid a deadlock, the holon-firm
that does not reach any agreement at the end of the nego-
tiation process is removed from the ranking list.
2.2. Centralised Approach (Benchmark)
In order to test the benefits of the adopted negotiation
approach for the HPS, we develop a centralized model
and a Mixed Integer Program (MIP). The objective func-
tion is the total profit maximization of the holonic net-
work as a whole:
max Profit(os)*
riceC tx
 (8)
being ij
the amount of capacity of the ith holon-firm
transacted with the jth holon-firm. Data knowledge of the
model is constituted by:
subject to the following constraints:
ij f
RCfor eachj
ij f
OCfor eachi
Constraint (9) forces the whole capacity transaction
toward the jth holon-firm to become lower than that re-
quested by the same holon-firm. Constraint (10) forces
the whole capacity transaction toward the ith holon-firm
to be lower than that offered by the same holon-firm.
3. Simulation Environment
The distributed network has been implemented and tested
through a simulation environment developed by using the
Java Development Kit (JDK) package. The adopted
modeling formalism is a collection of independent agents
interacting via a discrete events mechanism. The simu-
lation consists of a group of several interacting agents,
while simulation proceeds through alternative timetables
of discrete events. The considered schedules are data
structures that combine actions in the planned order. Two
kinds of packages have been developed: the holon-firms’
network and the statistical package.
The holon-firms’ network package consists of three
different agents with specific tasks:
the holon’s agent, supervises local holon-firm data
and algorithms;
the mediator agent, implements the negotiation me-
chanism among all the holon-firm agents involved
in the negotiation;
the scheduling agent, manages the discrete events
that determine the simulation runs.
The statistical package collects the data generated at
the end of each simulation run and generates both reports
and statistical analyses. The test environment considers
the following inputs:
holon-firms’ capacity: the capacity assigned to the
holon-firm at the beginning of each sub-period;
capacity demand: the product demand that the ho-
lon-firm tries to satisfy;
reconfiguration cost: the cost of the holon-firm to
change the production from one typology product to
number of product-type: the number of lines of
products requested by the market.
The simulations were conducted considering: a num-
ber (I) of six of holon-firms (I = 6) and a number (N) of
twelve sub-periods (N = 12). At the beginning of each
sub-period, the capacity re-allocation process is activated.
Input data were randomly generated, being randomly
drawn from different distributions as reported in Table 1.
Low and High are referred to the standard deviation’s
size of the normal distributions. The target profit value is
fixed for all the considered holon-firms. The simulation
has been conducted for different numbers of product-
types (k): k = 1, k = 5 and k = 10, in order to investigate
the behavior of the network when the number of pro-
Table 1. Input data.
Low High
Holon-firms’ capacity N(100,10%) N(100,50%)
Capacity demand N(100,10%) N(100,50%)
Reconfiguration costs N(2,10%) N(2,50%)
The Holonic Production System: A Multi Agent Simulation Approach
Copyright © 2010 SciRes. iB
ducts typologies change. The capacity costs are 10 units,
therefore the holon-firm reconfiguration costs are con-
sidered as 20% of the capacity costs. The holon-firms
have the same mark-up target: 20%. For each simulation
run, the goal was to evaluate the unallocated capacity
(UC) and the total amount of profit generated by the
holon-firms. Simulation experiments were conducted to
evaluate, for each run, the measures of performance with
a confidence interval of 95%. Combining different levels
of all four parameters, 24 simulation classes of experi-
ments were obtained as described in Table 2.
The following performance measures have been con-
the total profit (profit); given by the profits gene-
rated by all the holon-firms belonging to the net-
the unallocated capacity (UC); given by the sum of
the holon-firms’ unallocated capacities.
4. Simulation Results
Tables 3-6 report, in percentage, the difference between
the performances, evaluated using the proposed distrib-
uted coordination model in different environmental con-
ditions. In particular, Table 3 reports the results of the
Table 2. Experimental classes.
tions costs
type K
1 L L L 1
2 L L H 1
3 L H L 1
4 L H H 1
5 H L L 1
6 H L H 1
7 H H L 1
8 H H H 1
9 L L L 5
10 L L H 5
11 L H L 5
12 L H H 5
13 H L L 5
14 H L H 5
15 H H L 5
16 H H H 5
17 L L L 10
18 L L H 10
19 L H L 10
20 L H H 10
21 H L L 10
22 H L H 10
23 H H L 10
24 H H H 10
performance measured considering different degrees de-
mand variability.
Demand variability leads to a significant increase of
the unsatisfied demand, while the profit of the network
shows a small decrease. Table 4 reports the simulation
results when considering the distribution of the different
capacities possessed by the holon-firms in the network.
The distribution of capacity among the holon-firms
combined with network variability leads to high levels of
unsatisfied demand, while the total profit of the network
shows a low variability. However, the influence on the
performances of the network engendered by the holon-
firms’ capacity variability is not as strong as that caused
by customer demand variability.
Table 5 reports the simulation results considering the
distribution of reconfiguration costs among holon-firms.
The results of the simulation show that the distribution of
reconfiguration costs has a very low effect on perform-
ance measures.
Finally, the number of product types has been investi-
gated in order to understand how diversification affects
the performance of the system. Table 6 reports the in-
Table 3. Simulation resultscustomer demand.
Customer demand Demand unsatisfied Total profit
Low variability 402.4 13523.49
High variability 794.08 12855.83
Difference 97.34% 4.94%
Table 4. Simulation resultsholon-firm’s capacity.
Holon-firms’ capacity Demand unsatisfied Total profit
Low variability 427.4 13528.98
High variability 769.08 12850.34
Difference 79.94% 5.02%
Table 5. Simulation resultsreconfigurations costs.
Reconfiguration costs Demand unsatisfied Total profit
Low variability 584.53 13218.55
High variability 583.02 13218.55
Percentage difference 0.26% 0.00%
Table 6. Simulation resultsproducts typology.
Number of product typology KDemand unsatisfied Total profit
1 511.45 13362.2
5 616.81 13158.84
10 623.07 13134.21
Percentage difference 5-1 20.60% 1.52%
Percentage difference 10-1 21.82% 1.71%
Percentage difference 10-5 1.01% 0.19%
The Holonic Production System: A Multi Agent Simulation Approach
Copyright © 2010 SciRes. iB
K = 1 K = 5 K = 10 K = 1 K = 5 K = 10
K = 1 K = 5 K = 10 K = 1 K = 5 K = 10
Figure 3. HPS responsiveness to demand and capacity variability: Bar charts represent unallocated capacity and whole HPS
profit at different K diversification levels.
fluence of the number of products types k on the system
performance indicators, here calculated as the mean of
the customer demand and holon-firms capacity vari-
ability. The simulation shows that, while the unsatisfied
demand grows as the number of products types increases,
the effect on the network’s profit reduction is attenuated.
As a consequence of product differentiation, the respon-
siveness of HPS to market turbulence is able to alleviate
profit variations. Moreover, the difference in percentage
obtained among one and five product types is close to the
difference obtained with one to ten types. No relevant
percentage difference exists among the result with five to
ten products types. This last result highlights that, in a six
holons production system, the greatest profit make up
takes place with a products differentiation lower than
five typologies, and that product differentiation is a key
factor of adaptability to environment evolutions.
Figure 3 shows the bar charts, for each class of ex-
periments, representing the holonic system performance
variations due to variations of k (differentiation), meas-
ured by the HPS unallocated capacity and profit. The
four diagrams highlight that unallocated capacity raises
with k, whereas the degree of profit reduction is de-
creased. This counterintuitive result draws attention to
the mechanism of profit reduction attenuation, which is
typical of the holonic system. In fact, the negotiation
protocol combined with the holonic mediator function,
guarantees an optimal capacity allocation and distribu-
tion. While negotiated capacities are transacted at a lower
marginal profit, they still generate a supplementary profit
for the HPS as a whole.
Although the positive effects of the collaborative ne-
gotiation approach is clear for the holonic network per-
formance as a whole, the sustainability of the HPS de-
pends on the wealth of the single holons, upon which the
network structure relies. Further research development
could concern the analysis of the study of single holon-
firm profit dynamics, in order to understand the sustain-
ability of the multi-agent negotiation approach applied to
HPS in greater depth.
5. Conclusions
This paper proposes a cooperative approach among
holon-firms that operate in the same heterachical network
as holarchy. Cooperation is considered in terms of trad-
The Holonic Production System: A Multi Agent Simulation Approach
Copyright © 2010 SciRes. iB
ing capacity and the holon-firms can be sellers or buyers,
depending on the situation. The methods used to sup-
portthe holon-firms network are: a multi-agent system
(MAS) to implement the distributed architecture and a
negotiation approach to develop the cooperation mecha-
In this paper we use a simulation environment devel-
oped in Java language, in order to test the proposed ap-
proach. The performance measures considered have been
network profit and unsatisfied demand. These perform-
ances have been evaluated in several environmental con-
ditions: different predictability of customer demand, di-
verse holon-firms’ capacity and reconfiguration costs
variability in the network. The simulation results high-
lighted the added value produced by the cooperative ap-
proach of the Holonic Production System (HPS), espe-
cially in turbulent or variable environmental conditions.
the cooperative approach proposed is robust, in fact
the total profit of the network keeps the same level
in the different conditions tested;
the variability of reconfiguration costs among the
holon-firms has very low effects on the performance
the increase of the other three parameters (demand
variability, holon-firms’ capacity variability and
number of product types) leads to an increase in the
level of unsatisfied demand;
the variables that influence HPSs’ performance are,
in decreasing order: demand variability, variability
of holon-firm capacity and number of product types.
The analysis conducted highlighted that the adoption of
a cooperative approach leads to superior performance on
behalf of the holonic network when the environment is
more dynamic. We call this effect network resilience to
market turbulence threats.
Further developments of the research proposed could
concern: a dynamic investigation of the evolution of the
participants in the network, considering also the effect of
the growing number of holon firms and product diversi-
fication in the HPS; an analysis of the effects of the en-
vironmental unpredictability on the performance of the
single holon firms, in terms of demand satisfaction and
profit, as a complement to the system level analysis pro-
posed in this paper; furthermore, an extension of the
aforementioned study would consent to improve the basic
understanding of the local holon sustainability versus the
global network welfare.
The simulation environment has been entirely devel-
oped by using an open source code with an object oriented
architecture (Java), so that the multi agent framework
developed for our research purposes can be easily up-
graded to build a real digital managerial system for real
holon-firm networks. This choice allows to reduce in-
vestment time and risk of building a real system able to
manage networked holon-firms like a HPS.
[1] G. Dominici, “Demand Driven Supply Chain ed Innova-
zione: Il Sistema Logistico-Produttivo per la Soddisfazi-
one dei Bisogni del Cliente,” In: A. Purpura and G. Fazio,
Eds., Economia e Gestione Dell’innovazione nelle PMI,
FrancoAngeli, Milan, 2008.
[2] G. Dominici, “Holonic Production System to Obtain Flexi-
bility for Customer Satisfaction,” Journal of Service Sci-
ence & Management, Vol. 1, No. 3, 2008, pp. 251-254.
[3] F. Frederix, “From Production to a Product Perspective.
New Industrial Scenario?” In: S. Yoon, et al., Evolution
of Supply Chain Management, Symbiosis of Adaptive
Value Networks and ICT, Kluwer Academic Publishers,
Norwell, 2004.
[4] L. Gou, P. B. Luh and Y. Kyoya, “Holonic Manufac-
turing Scheduling: Architecture, Cooperation Mechanism,
and Implementation,” Computers in Industry, Vol. 37, No.
3, 1998, pp. 213-231.
[5] J. Hatvany, “Intelligence and Cooperation in Heterarchic
Manufacturing Systems,” Robotics & Computer-Integrat-
ed Manufacturing, Vol. 2, No. 2, 1985, pp. 101-104.
[6] H. van Brussel, L. Bongaerts, J. Wyns, P. Valckenaers
and T. van Ginderachter, “A Conceptual Framework for
Holonic Manufacturing: Identification of Manufacturing
Holons,” Journal of Manufacturing Systems, Vol. 18, No.
1, 1999, pp. 35-52.
[7] P. Valckenaers, E. Bonneville, H. van Brussel, L. Bon-
gaerts and J. Wyns, “Results of the Holonic Control Sys-
tem Benchmark at the KULeven,” Proceedings of Com-
puter Integrated Manufacturing and Automation Tech-
nology Conference, Troy, New York, 1994, pp. 128-133.
[8] T. Moyaux, B. Chaib-draa and S. D’Amours, “Supply
Chain Management and Multiagent Systems: An Over-
view,” In: B. Chaib-draa and J. P. Müller, Eds., Multi-
agent Based Supply Chain Management, Springer, On-
tario, 2006.
[9] M. Paolucci and R. Sacile, “Agent-Based Manufacturing
and Control Systems,” CRC Press, Boca Raton, 2005.
[10] G. M. Golinelli, “Viable System Approach (VSA),”
Community Economic Development Association of
Michigan, Michigan, 2010.
[11] A. Koestler, “The Ghost in the Machine,” Arkana, Lon-
don, 1967.
[12] C. Saccani, “Il Sistema Olonico,” Sistemi & Impresa, Vol.
2, 1996, pp. 29-45.
[13] M. A. Arbib, “The Handbook of Brain Theory and Neural
Networks,” MIT Press, Boston, 1995.
[14] P. Renna and P. Argoneto, “Capacity Allocation in Multi-
Site Environment: A Multi Agent Systems Approach,” In:
Proceedings of Intelligent Computation in Manufacturing
Engineering, Naples, 2008.