A Methodological Approach for Agent Based Simulation of Mass …ACL 05.2... · 2018-02-18 · associated with modelling supply chains in a mass customization context. In section three,

O. Labarthe, B. Espinasse, A. Ferrarini, B. Montreuil (2005), « A Methodological Approach for Agent Based Simulation of Mass Customizing Supply Chains », in: Journal of Decision Systems (JDS), vol. 14, n° 4, pp. 397 – 425.

A Methodological Approach for Agent Based Simulation of Mass Customizing Supply Chains Olivier Labarthe*, ** — Bernard Espinasse* — Alain Ferrarini* — Benoit Montreuil** * LSIS UMR CNRS 6168, Université Paul Cézanne, Domaine universitaire de St Jérôme, 13397 Marseille - Cedex 20, France. [email protected] [email protected] [email protected] ** Canada Research Chair in Enterprise Engineering, CENTOR Université Laval, Pavillon Palasis Prince, G1K 7P4, Qc, Québec, Canada. [email protected]

ABSTRACT. Mass customization has a significant impact on supply chains. The management of these highly dynamic organizations requires the development of adapted decision support systems. The agent approach appears fit for modelling and simulating mass customization supply chains. This paper presents a methodological approach for agent oriented modelling and simulation that is based on specific modelling levels and formalisms. The approach is illustrated through a case from the Golf industry. RÉSUMÉ. La production à grande échelle de produits personnalisés a un impact considérable sur les chaînes logistiques. La gestion de ces organisations fortement dynamiques, nécessite le développement d’outils d’aide à la décision adaptés. L’approche agent apparaît indiquée pour la modélisation de ces chaînes dynamiques et pour le développement d’outils de simulation. Ce papier présente une démarche méthodologique pour la modélisation et la simulation orientées agents reposant sur des niveaux de modélisation et des formalismes spécifiques. Cette démarche est illustrée sur un cas issu de l’industrie du Golf. KEYWORDS: Supply Chain, mass customization, agent-based modelling and simulation, methodological approach.

2 Journal of Decision Systems. Volume 13 – No. 1/2004

MOTS-CLÉS: Chaîne logistique, personnalisation de masse, modélisation et simulation orientées agents, démarche méthodologique.

A methodological approach for agent based simulation 3

1. Introduction

In order to respond to the global competitive pressures, enterprises collaborate ever more with their suppliers and clients, become more agile and develop mass customization capabilities. Relational transformations give rise to new forms of network organizations. Enterprises have increased their design and management scope to encompass their entire supply chain. The intensifying of information sharing transforms business processes. From a management and decision making perspective, this sets the stage for increased complexity and dynamics.

From an operational perspective, orchestrating the flow of personalized products through a supply chain involving multiple collaborating enterprises implies complex decision making in a highly stochastic fast-pace environment. From a tactical and strategic perspective, the network and its processes require continual adjustments in terms of their design and their management policies. The complex, stochastic dynamic and high impact nature of these issues requires careful studies prior to action. Yet such studies can rarely be based on pilot tests on the real system due to cost and time constraints. Simulation based studies appear much better suited to deal with these issues.

Yet current industrial system modelling and simulation suffer from weaknesses in their capability to apprehend the inherent constraints of network organizations. The enterprise modelling methods do not allow taking into account the agility of such organizations and the dynamic nature of their environments. Frameworks have been introduced for modelling manufacturing networks (Min et al., 2002), (van der Zee et al., 2005). They allow expressing the diversity of structural configurations and the decentralization of network components.

This paper presents an agent oriented methodological approach for modelling and simulation of mass customizing supply chains. This approach is to allow the development of simulation based decision support systems enabling smart management of such highly dynamic chains. The agent approach, inspired from distributed artificial intelligence, is deemed indicated for the study of the dynamics interactions and the behavioural representation of supply chain entities, and to take into account the dynamic and stochastic nature of their environment.

The paper is structured as follows. In section two, we develop the problem associated with modelling supply chains in a mass customization context. In section three, is proposed a methodological approach for agent oriented modelling and simulation of dynamic supply chains. This approach is based on a multi-level modelling framework: domain, conceptual agent and operational agent. A strong emphasis on domain modelling characterizes the first step of the proposed approach. The section four presents the second step, the conceptual agent modelling, and develops the proposed formalisms and the rules for deriving the domain model. In section five, as a third step, are developed the operational agent modelling, its formalisms and the rules for deriving it from the conceptual agent model. Finally section six concludes and presents perspectives for further research.


2. From mass customization to agent oriented simulation

Large scale production and distribution of personalized products has a significant impact on supply chains, both from design and management perspectives. There is a lack of methodological tools for analyzing the links between the product characteristics and the design and management of the supply chain.

In this section we first introduce mass customization, leading supply chains to operate in a highly dynamic and stochastic context. Then we survey current approaches for modelling supply chains and state the specific issues associated to modelling dynamic chains. Then we discuss the interest of the simulation approach for studying dynamic chains, and the characteristics of agent oriented simulation.

2.1. Mass customization

The evolution of supply chains traces a redefinition of the positioning of enterprises relative to the demand. The development of differentiating capabilities has lead enterprises to develop an offer for highly parameterized products delivered faster and faster (Childerhouse et al., 2000). The understanding and anticipation of customer expectations allow enterprises to transform their value creation processes. To better satisfy the demand for personalized products, industrial organizations have more and more adopted mass customization (Pine 93). Consumer centric organizations are subject to highly dynamic and stochastic demand due to the massive distribution of personalized and tailored products. This leads to the definition of product/market segments which condition flow streams through existing networks and new networks. The added requirements for a continuous flow of innovation have render ever more complex the management of the demand for personalized products. In this highly dynamic context, classical demand forecasting models are limited due to the lack of data and history (Fisher 97).

The association of the stochastic nature of demand, the heterogeneity of markets and the reduction of order-to-delivery lead times, reinforces the complex dynamics associated to product personalizing. These factors make it ever more difficult to elaborate management policies able to anticipate the product transformation activities needed to satisfy demand. The flow of products, the sourcing and supply diversity, the dimensioning of storage units, the agility of manufacturing systems, and the delivery speed accentuate the difficulties for optimizing the physical flow of ever more innovative products. Enterprises evolve in a context where interactions and decision making become subject to the volatility of consumer expectations.

2.2. Modelling dynamic supply chains

Modelling a supply chain consists in describing its workings so as to study them with the aim to improve its performances. Analysing the models and the associated experimental results allows enterprises to evaluate the agility of their organizations


and to anticipate their reactions in face of the environment dynamics (competition, etc.). This evaluation is an element of an approach by enterprises for understanding and improving their value creation processes.

Models, methods and tools developed for enterprise modelling (Vernadat 02), are widely used to represent manufacturing and logistics systems, (Bussmann et al., 2004). However the models extracted from enterprise modelling do not allow to describe all the specific characteristics of a chain and to consider this chain as an enterprise and not as a network of enterprises. Thus proprieties such as the nature of the organizational structure, the autonomy of cooperating entities, the interaction dynamics, as well as the independence of control processes are generally not expressed or made explicit in models generated using the enterprise modelling field.

Analytical models of supply chains represent them from a quantitative perspective. Multiple approaches can be distinguished (Parunak et al., 1999). Control theory represents the dynamic of the system, (Disney et al., 2002), through time based on behavioural linearization hypotheses. Thus resulting models require adaptations of the real world conditions and do not allow representing the entirety of the supply chain. The Operations Research approach, (Goetschalckx 00) (Lakhal et al., 2001), requires statistical approximations and is not directly based on a temporal scale. Thus the dynamic characteristics of the system are difficult to apprehend. The complexity inherent to modelling supply chains requires the elaboration of simplifying hypotheses which are often highly restrictive. A number of these limitations can be overcome through modelling approaches relying on simulation.

In the Simulation approach, the representation of supply chains operating in highly dynamic and stochastic environments requires the elaboration of adapted modelling capabilities. In order to analyse a chain and study its behaviour, it is necessary to be able to represent its structure and its dynamics so as to enable its simulation (Gunasekaran 04). This generates problems linked to the exploitation of current simulation tools as we will approach in the next subsection.

2.3. Agent oriented simulation

Simulation relies on the exploitation of a set of models and methods allowing to approach and to imitate the behaviour of a real physical system. The majority of supply chain simulation works are based on discrete event models. This type of model insures that the temporal component is taken into account and it favours the understanding of the behaviour of the supply chain in its environment.

Modelling supply chain for the design and the implementation of discrete event simulation can be based on different approaches. Research works presented in (Biswas et al., 2004) and (van der Zee et al., 2005) elaborate object-oriented model of supply chain. Another computerized technique is related to agent oriented approaches. Concepts such as autonomy, mental states, communication through language act, as well as the organization into societies of interacting entities, are notions which differentiate the agent and object concepts (Wooldridge et al., 2001).


Agent oriented simulation leads to the observation through time of as system and its constituting entities. This approach is first based on individual modelling, that is the identification of individuals and the representation of their behaviour and organization, (Fox et al., 2000). Second, it is based on the definition of observables which are quantifiable properties characterizing one of several individuals. These observables have values which are transformed through the actions realized by individuals. The state of the system then results from the behaviour of individuals and their interactions, (Sadeh et al., 2003). The transposition form individual to agent allows developing agent oriented systems which are amenable to implementation, enabling simulation for studying the dynamic behaviour of the modeled system, (Brueckner et al., 2005). However the design of agent oriented simulations requires enacting specific knowledge and competencies.

A number of problems arise when attempting to use agent oriented simulation. Agent oriented simulation is concurrently a software product, an agent based software product and a simulation tool. As a software product, it necessitates the elaboration of classical software engineering approaches guiding its development process, insuring its quality, its robustness and its evolution capability. As an agent based software product, it relies on development methodologies which are mainly adaptations or extensions from knowledge oriented or object oriented methodologies, Unified Modeling Language (Booch et al., 1999), etc. Other approaches integrate directly the properties of agents into the core of the methodology and are guided by the type of agent architecture and organization, which is detailed more finely at implementation phase, (Wooldridge et al., 2001).

As a simulation tool, it raises the problem of the design of the agent oriented executable model and of its implementation enabling simulation experiments. The complexity associated with the design and implementation is expressed in (Fishwick 98) by classifies the elements defining the simulation field using three levels: Model Design, Model Execution and Execution Analysis.

2.4. Conclusion

The mass customization context induces an environmental dynamic characterized by a stochastic demand in terms of consumers, products and markets. The supply chains put in place to enable mass customization are characterized by fast paces dynamic interactions requiring adapted modelling capabilities.

Agent oriented modelling and simulation enable to study the behaviour of complex dynamic and distributed systems, such as supply chains as proven through multiple research endeavours. The design of agent oriented simulation models is already relatively well mapped, mainly through the proposition of methods derived for software engineering. Agent oriented modelling and simulation of supply chains in general, and dynamic mass customization chains in particular, must be supported by an adapted domain modelling due to their complexity and particularities. Defining such a specific methodological approach appears necessary.


3. A methodological approach for agent oriented modelling and simulation of supply chains in a mass customization context

Realistic modelling of a mass customizing supply chain relies on the representation of its structure and dynamics. The dynamics is mostly characterized by: i) the decentralization of the decisional processes, ii) the autonomy of cooperating entities, and iii) the distributed nature of interactions. Taking into account the structural and dynamic aspect in modelling and simulation processes, necessitates a methodological framework ensuring a continuum from developing domain models to implementing and executing them.

This section presents the structural specificities of supply chains in a mass customization context. A reference framework for such supply chains is proposed. An industrial case of mass customization is presented to illustrate the modelling process. At last this section shows the proposed agent oriented methodological approach, with its modelling framework structured on several modelling levels.

3.1. Structure of mass customizing supply chains

Market evolution, characterized by a dynamical change of the needs, constrains decision makers to manage innovating products which are hardly foreseeable. The execution of a mass customization strategy is related to the management of innovative products, as shown in studies presented in (Duray 02).

The personalization level describes the relation between the customer and the supply chain, related to the delivery time and the response time. Deep personalizing is much tougher to reliably and recurrently deliver fast at minimal cost than cosmetic personalizing. This influences the managing policy in order to deliver the personalized offers within a satisfying response time.

In the case of a demand driven control mode, the pull mode, the product flow is triggered by orders (to Order). In the push mode, products are pushed toward the downstream supply chain. This last case generates product inventory due to the manufacturing lead times or due to economy of scale. The needs expression induces a contact point between customer and supply chain; this point which is function of the personalization level represents the order penetration point (Hoover et al., 2001).

A reference framework of the supply chain, presented in figure 1, classifies the activities according to eight personalization levels (Labarthe et al., 2003). Each actor of the chain is characterized according to its competences, roles and responsibilities, and has to coordinate its activities with the other actors to answer the demand. These characteristics define an activity network. The issue of an order is represented on the network by the order penetration point corresponding to a personalization level. The order penetration point determines the entrance of the order in the product flow. The identification of the personalization level allows defining the decoupling point on the actors' network. This point has an impact on inventory management and on relations to establish among the actors situated upstream and downstream this point.


Figure 1. Personalization levels and supply chain coordination

3.2. Supply Chain Modelling by responsibility networks

3.2.1. Responsibility networks

We base our approach on the responsibility networks for modelling the structural aspects of supply chains. As introduced in (Montreuil et al., 1996), responsibility networks formalize the role(s) of each element of the network. This(ese) role(s) define the activities for which each actor is responsible and the interacting client-supplier relationships between actors. In our approach, these responsibility networks are defined according to an organizational-level resolution: inter-enterprise, inter department, and so on. The inter-enterprise responsibility network can be broken down in other inter-department responsibility networks, and so on.

To model the dynamic aspects of supply chains, we were inspired by the modelling framework proposed in the NetMAN project (Montreuil et al., 2000, and Frayret et al., 2001). This framework deals with the design and management of manufacturing networks, be they hierarchical or heterarchical. In a dynamical environment this framework allows the planning, the control and the management of the activities of the actors of the network. We extend this framework to better take into account the specificity of the studied supply chains. Thus, the Domain Model of Supply Chains is composed of two complementary models:

The Structural Model describes the structure of the supply chain, presents its actors and the material flows which exist among these actors. The Dynamic Model defines the behaviour of each actor and clarifies the interaction modes and the coordination modes used, in particular it states the nature of exchanged information.

The following sub-section details this domain modelling of supply chains on a golf industry case study.

S P A G F E D S C

Supplier Manufacturer Manufacturer Distribution Center

Retailer Demand

ActorsNetwork

Supplying ProducingAct ivity Network

Order Penetration Point

PersonalizationLevels

Assembling Grouping Finishing Packaging Distributing Sales Buying

Leve

l VII

Leve

l VI

Leve

l V

Leve

l IV

Leve

l III

Lev e

l II

L eve

l I

L eve

l 0

DP

DP

DP

DP

DP

DP

DP

DP Design To Order

Buy To Order

Make To Order

Assemble To Order

Group To Order

Finish To Order

Package To Order

Deliver To Order

Customer s Relations

Push Flow

Pull Flow

Transport Physical Flow Order DP Decoupling Point S Supplier C Customer


3.2.2. Domain modelling by responsibility networks : the golf club industry case

We illustrate the proposed approach through the supply chain of a golf club manufacturer operating as a mass customizer, (Montreuil et al., 2005). The characteristics of this industrial field are illustrated figure 2 (Labarthe et al., 2004). This field is characterized by a demand which is function of the personalization levels offered by the manufacturer in each market zone.

Figure 2. Supply chain of an illustrative mass customizing golf club manufacturer

Each actor is considered as an independent company. This allows to propose in figure 3 the responsibility network of the Golf Club Manufacturer at the inter-enterprise level. The responsibility network is composed of a set of suppliers having as responsibility to supply the Assembler. It is the responsibility of these manufacturers (Assembler, Processor and Fulfiller) to transform raw materials and / or components into final products. Once these transformations are accomplished, final products are delivered to Distributors. The distribution centers deliver the finished products to Retailers which have the responsibility to propose and sell the products to the customers. The Structural Model represents the set of responsibility networks for a specific level of abstraction. Figure 4 illustrate the decomposition principle of the Golf Club Manufacturer at the inter-department level.

The Dynamic Model describes the behaviour of the Structural Model elements. All the business units, actors, are represented by Centers in the Dynamic Model, according to the NetMan framework. These centers coordinate by information flow interactions with centers and external actors of the considered organization. The information flows used to express the dynamics are: i) the needs, ii) the offers, iii) coordination data, and iv) coordination models. The needs are represented by orders characterized by the personalization level. Offers are expressed according to personalization levels and associated delivery times (Poulin et al., 2006).

x 50 000x 300

x 4 Markets Zones

x 1 European Market

x 50 000x 500

x 1 US Market

Molds Sub -Contractor

Molded Heads Supplier

Shafts Supplier

Grips Supplier

Forged Heads Supplier

Finished Products Distributor

Customized Golf Club Manufacturer x 4 Markets Zones

DEMAND

Golf Club Manufacturer

Golf Club Sub -Contractor

Others MarketsSupplier Manufacturer Distributor Retailer CustomerPhysical

Flow

Informational Flow


The behaviours related to the coordination modes are described according to two approaches. Network models describe the actors' coordination inside the network while center models are focused on the internal coordination of a supply chain actor's activities. These models have upstream and downstream transmission modes. Figure 5 shows the Dynamic Model of the golf club mass customizer case.

Figure 3. Responsibility network of the Golf Club manufacturer

Figure 4. Structural Model of the Assembler at the inter-department level

Figure 5. Inter-Enterprise Dynamic Model of the Golf Club Manufacturer

DPDP

NetMan CenterNetMan CenterModels

Material Flows

Needs Expression

Information Sharing

Raw Material

Work In Process

Final Product

De-coupling Point

Handling and transport

Business Process

A

Models Transmission

Demand Models

Fulfiller (Finishing, Packaging)

F EAssembler (Producing, AssemblingAssembling, Grouping)

A

DPDP

GP

Models Periodical Update

EFCenters Models

Network Models

Distributor ModelFulfiller Model

Assembler Model

A G

DPDP

CustomerRetailer Model Demand Model

E

E

A GPAssembler

TS S S E

Golf Club Manufacturer

PS S TProducer

PS S TProducer

PS S TProducer

PS S TProducerMolded Heads

Shafts

Grips

Forged Heads

PProcessor

S S TGolf Club Sub-

Contractor

PProcessor

S S TGolf Club Sub-

Contractor

F EFulfiller

S TSCustomized

Golf Club

DS TFinished Products

Distributor

Input products Output ProductsSUPPLIERS

CLIENTS

Actor DemandPhysical Flow

x n Entity NumberCustomer

Processor ProducerAssemblerFulfiller

DistributorRetailer

Business Process (Assembling, Stocking, Finishing, Producing, Transport, Grouping, Sales , Packaging )

A

EV

S S

DS TDistributor

PS S TProducer

A

S TDistributor

G

AssemblerS S TDS T

DistributorS TE

DistributorDS T

Distributor

Molded Heads Producer

Forged Heads Producer

Finished Products

Golf Club Sub-Contractor

Customized Golf Club

Golf Club Manufacturer (Assembler)

DS TDistributor

DS TDistributor

DS TDistributor

Golf Club Sub-Contractor

DS TDistributor

DS TDistributor

Grips Producer

DS TDistributor

Shafts Producer

SUPPLIERS

CLIENTS


3.3. Methodological approach for agent oriented simulation design

We here introduce a methodological approach for implementing agent oriented simulations of mass customizing supply chains. The proposed approach is composed of several stages ensuring a continuum in the expression of models from domain models to their transposition into agent models leading toward implementation in a simulation tool. The approach exploits the responsibility networks as described in the previous section. This allows a fine representation of the structural and dynamic aspects of supply chains into a domain model.

This section is structured as follows. Next the different modelling levels are presented, as well as the associated specific models and the participants to the realization process. Then the different stages of the method, relying on these levels, are detailed. Finally the associated main design processes are presented.

3.3.1. Modelling levels

The proposed methodological framework relies on the distinction between four interrelated modelling levels, each one having its own concern:

The conceptual domain level aims to formalize through a domain model the pertinent knowledge on the studied supply chain in a way domain experts can readily manipulate. The point is to describe the structure of the manufacturing and supply network, and the behaviours of the supply chain actors. This level requires formalisms from the manufacturing network and supply chain field.

The conceptual agent level develops an agent oriented modelling of the system, starting from the domain model, disregarding technical considerations associated with the execution of any simulation. This conceptual agent modelling relies on several sub-models that together specify the architecture of the Multi-Agent System (MAS). This approach also specifies for each agent its behaviours as well as its interactions and their nature.

The operational agent level develops an agent oriented operational modelling, in which each agent identified at the conceptual level is specified and implemented. At this operational level the specificities and the constraints related to the agent development environment, are taken into account; these constraints can be based on the agents internal architecture, whether of cognitive nature, whether of reactive nature or even hybrid. Also, the agents' beliefs and behaviours are specified.

The exploitation level is concerned with the implementation of the operational model inside a computer system. The technical constraints ignored at the operational level are taken into account. The Multi Agent System is deployed inside a software environment, allowing its execution for running simulations.

3.3.2. The three stages of the methodology

The proposed methodological approach recommends three stages positioned on the previously presented modelling levels. The first stage relies on the two first modelling levels; the two following stages are each based on each of the following


modelling levels. It can be noticed that this approach is quite similar to the approach proposed in (Drogoul et al., 2002). The stages of the approach are:

Conceptual Modelling: this stage, which consists in modelling the structural and dynamic aspects of supply chain, is realized in two steps: first an adapted domain model is generated, and second an agent oriented conceptual modelling is derived and crystallized in the Conceptual Agent Model (CAM).

Domain modelling is mostly focused on the supply chain components and on the nature of their interactions. It has already been detailed and illustrated through the golf club mass customizer case in previous sub-section. It mainly relies on the responsibility networks. The obtained model is the Domain Model (DM).

The agent paradigm is then used to enrich this domain model by the description of its components' decisional processes, from a social and local point of view. The agent oriented modelling allows to represent the dynamic characteristics of the system through the actions and behaviours of its interacting entities. The model resulting of this stage is named the Conceptual Agent Model.

Operational Modelling: this stage consists in deriving, from the previous CAM, a Multi-Agent model to simulate the physical activities, the decisional processes, and the informational and material flows of the supply chain in a mass customization context. The resulting model is the Operational Agent Model (OAM).

Experimentation: the result of this stage is constituted of the Multi Agent System and of the experimental framework. From the previous OAM, the MAS is designed. The resulting MAS is then integrated inside a simulation environment enabling simulated experiments on the supply chain.

3.3.3. Participants in the modelling process

As specified in (Drogoul et al., 2002), three different types of participants collaborate in the development of a simulation tool: the domain expert, the modeller expert, the computer scientist. These participants intervene in the course of one or several stages of the methodological approach presented in this paper, contributing in one or several modelling levels:

The domain expert, defines the needs for simulation and takes on the description of the system by the design of the DM. His privileged intervention level is the domain level.

The modeller expert transforms the domain model designed by the domain expert into a conceptual agent model. He bases this transformation on the agent paradigm. He intervenes essentially at the conceptual agent level.

The computer scientist is interested in specifying the operational agent model, leading to the design of software agents then implemented in a simulation environment. His privileged intervention level is the operational level.

The domain expert and the modeller expert interact to ensure a valid transposition of the domain model into a conceptual agent model. The modeller


expert and the computer scientist similarly interact to ensure a valid transposition of the conceptual agent model into an operational agent model. Once the simulation model is implemented, an experimental expert, the user, enters into play to conduct and statistically analyze the simulation experiments. He interacts with the domain expert, focusing on the fulfilling the needs for simulation. He interacts with the modeller expert and the computer scientist when experiments are planned that require not yet modeled or implemented features.

3.3.4. Main design processes

Now that have been defined the goals and deliverables associated to each stage, we focus on describing the main design processes leading to valid models : i) the abstraction process, ii) the agentification and validation processes, iii) the specification and verification processes, and at last iv) the implementation and execution processes.

Abstraction process: the domain expert represents the real system and specifies the expected experimentations by the definition of the phenomena to be observed. These elements are derived from macro-knowledge and micro-knowledge, as defined in (Drogoul et al., 2002).

The micro-knowledge results from observations on the real system, made by the domain expert. These observations translate his view of the system elements and of their organization. The abstraction process translates this view on the basis of a modelling framework related to the domain of the study. It thus facilitates the description process realized by the domain expert.

The macro-knowledge concerns the hypotheses that the domain expert wants to test through the experiments. The elements which allow expressing these hypotheses are: simulation needs, observables, and simulation scenario. The needs are used to delimit the extent and the level of detail needed for designing the DM. The observables allow the domain expert to express the properties to be observed, and this is possible as soon as the CAM design stage. The scenarios, which intervene at the level of the exploitation of the simulation tool, allow configuring the parameters for simulation control.

Agentification and validation processes: the agentification process consists in translating the DM into a conceptual model based on the agent paradigm: the CAM. This stage allows representing the entities composing the DM with agents or objects, and allows to define the interactions which connect them. This CAM defines a Multi Agent organization on the basis of the observables, pointing out the behaviours to be simulated. This conceptual modelling necessitates several iterations to obtain a model validated according to the simulation needs. The validation process involves both the modeller expert and the domain expert.

Specification and verification processes: the CAM is the input to the operational modelling stage. The specification process consists in specifying the MAS from the CAM expression. The MAS constitutes the Multi Agent simulation model, also named the operational agent model (OAM). This specification requires


the use of formalisms adapted to the representation of software agents' behaviour, of their mental state, and of their interaction modes. The OAM constitutes an agent oriented software design solution of the CAM. A verification process, led jointly by the computer scientist and the modeller expert, must allow ensuring the reliability of the OAM relative to the CAM.

Implementation and execution processes: from the OAM, the computer scientist implements the Multi Agent System previously specified. This implementation stage necessitates numerous tests to ensure that the MAS corresponds to the simulation needs. The MAS is then integrated inside a simulation environment and can be used to simulate the real system on various scenarios. This constitutes the execution process. This process is concerned with the parameterization of the scenario and the display of results.

The different models, steps, transition phases and actors associated to this methodological approach are presented in figure 6.

Figure 6. Methodological approach for agent oriented modelling and simulation

The following sections detail the development processes of the CAM and OAM, and, in particular, the derivation rules used respectively from the DM and the CAM during the agentification and specification phases.


4. From domain modelling to conceptual agent modelling

This section concerns the Conceptual Agent Model elaboration process. The Domain Model (DM) is represented as a multi-agent organization (the agentification of the DM). The organization is transformed according to the simulation objectives (individual or group behaviours to observe during the simulation).

4.1. General agentification process of the Domain Model

The agentification process is based on a differentiation concerning two agent types at the conceptual level. The Actor-Agent represents a center according to the NetMan framework and is defined as an actor of the supply chain which has one or several roles (a set of competencies). The Activity-Agent represents an activity performed by an Actor-Agent as a business process at the DM level. For the agent conceptual modelling specific representation formalism is proposed (cf. figure 7).

Figure 7. The formalisms adopted at the domain and conceptual levels

The agentification process has to permit the structural coherence between the DM and the CAM. It is composed of the following modelling tasks:

1. Each actor of the DM is represented by an Actor-Agent in the CAM. 2. The physical flows of the DM are represented and their nature is defined. 3. The informational flows between the actors of the DM are represented in the

CAM and their nature is defined. 4. The organizational boundaries of each Actor-Agent are defined. 5. Each activity of an actor of the DM is represented by an Activity-Agent

associated to the Actor-Agent (inside its organizational boundaries). 6. The physical flows between the activities of the DM are represented in the

CAM, for each flow the products and the information are specified. 7. The informational flows between the activities of the DM are represented in

the CAM, for each flow the products and the information are specified. Each of these modelling tasks of the agentification process is detailed in the

following sub sections.


4.1.1. From the NetMan centers to the Actor-Agents

From the DM, according the agent paradigm, this modelling task defines a first population of agents composing the CAM. Each NetMan center defines an Actor-Agent in the CAM (cf. figure 8).

Figure 8. Defining the Actor-Agents of the CAM

4.1.2. Representation of the physical interactions between Actor-Agents

Each physical flow between actors of the DM is represented by a physical interaction between Actor-Agents. To represent in the CAM a product that moves as a physical flow between two Actor-Agents, an object is introduced and specified on the arrow representing the physical interaction. Figure 9 represents that center A supplies product X to center B, both at the domain and conceptual agent levels.

Figure 9. Physical interaction representation between Actor-Agents

4.1.3. Representation of the informational interactions between Actor-Agents

This modelling task translates all the informational flows between actors of the DM into informational interactions between Actor-Agents. Each informational flow of the DM is identified and specified in the CAM by a number on the interaction arrow. At this conceptual level, several cases are possible. Figure 10 illustrates two of these cases. In the first example, the informational interaction is simple: center B transmits its needs to center A, as shown using a simple arrow. The second example illustrates an informational interaction resulting of a needs expression (Center B to center A) and also a sharing of models between the two centers (double arrow).

Center A Center B

EP

E

P

Center A Agent Center B Agent

Domain Model Conceptual Agent Model

Center A Center B

EP

E

P

Center A Agent Center B Agent


X


Figure 10. Informational interactions representation between Actor-Agents

4.1.4. Organizational boundaries of Actor-Agents

This modelling task defines the organizational boundary of each Actor-Agent. Such boundaries are specified as in figure 11 by a line surrounding the agent symbol. The physical interactions between two agents are specified between the boundaries and not between the two agent symbols.

Figure 11. Representation of Organizational boundaries of Actor-Agents

4.1.5. From NetMan center activities to Activity-Agents

In this modelling task, each activity of a center in the DM is represented in the CAM by an Activity-Agent linked to the Actor-Agent associated to this center. This Activity-Agent is specified inside the organizational boundary of the concerned Actor-Agent. In figure 12, the activities of center A are represented by four Activity-Agents in the organisational boundary of the Actor-Agent Center A.

Center A Center B

EP Center A Agent Center B Agent


X

Center A Center B

EP

E

P

Center A Agent Center B AgentX

Exemple 1

Exemple 2

1 : Needs Expression (Center B Agent to Center A Agent)2 : Models sharing (Center B Agent to Center A Agent)3 : Models sharing (Center A Agent to Center B Agent)

1

1, 2, 3


Figure 12. Activities representation at the conceptual level

4.1.6. Physical interactions between Activity-Agents

Physical flows between activities are not specified at the domain level in the DM. Precedence relationships between activities permits to define the dynamics of the product processing. These physical flows between activities are in addition of the physical flows exchanged between actors. In the CAM, these physical flows between activities are represented as interactions between Activity-Agents according to their precedence relationships. Such interaction is specified by a simple arrow, either between two Activity-Agents (according to their precedence relationships) of a same Actor-Agent (cf. activities a1 and a2 in figure 13), or between two Activity-Agents of two different Actor-Agents (cf. activities a1 and b1 in Figure 13). Each product moving through the physical flow is represented by a named object specified on the arrow of interaction (cf. objects U and X in Figure 13).

Figure 13. Representation of physical interactions between Activity-Agents


4.1.7. Informational interactions between Activity-Agents

The representation of informational interactions between Actor-Agents and Activity-Agents is done in two steps:

− First, the interactions between Actor-Agents and Activity-Agents have to be defined. At this modelling level, several cases are possible: i) the informational links are defined between an Actor-Agent and all its Activity-Agents, ii) the informational links are defined between an Actor-Agent and all its Activity-Agents, but also between each Activity-Agents according to the precedence relationships in the transformation process, iii) the informational links are defined between the Actor-Agent and all Activity-Agents, but these links are not based on the precedence relationships of the transformation process (according defined observables), and iv) the informational links are defined between the Actor-Agent and some of its Activity-Agents (according defined observables).

− Then, the nature of these interactions has to be defined. As for informational interactions between Actor-Agents, the content and the direction of the informational exchange are specified.

In general, an Actor-Agent establishes a bidirectional informational flow with the set of its Activity-Agents. Figure 14 illustrates case (iii) introduced above.

Figure 14. Informational interactions representation in the multi-agent organization


4.2. Conceptual agent modelling of the Golf case study

To illustrate the Conceptual Agent Modelling of the golf club mass customizer, we focus here on two actors of the supply chain: the Customized Golf Club Manufacturer (Fulfiller) and the Finished Products Distributor. Their interactions are represented by elementary informational and physical exchanges. The specification phase allows us to formalize the DM using the agent paradigm. As one possible illustrative modelling alternative, the Customized Golf Club Manufacturer is represented in the CAM by a set of five Agents. As shown in Figure 15, four agents are respectively responsible for each of the main activities actually touching the products: Supply Golf Club, Customize Golf Club, Pack Golf Club and Deliver Golf Club. The fifth Actor-Agent is the Fulfiller Agent, responsible for managing demand and planning personalization process. Each Actor-Agent and Activity-Agent is characterized by its role, its responsibilities and by the activities it is able to realize.

Figure 15. Conceptual Agent Model of the Fulfiller-Distributor interactions

5. From conceptual agent modelling to the operational agent modelling

The agent operational modelling develops in the Operational Agent Model (OAM) an implementation solution of the CAM for a specific multi-agent environment. The OAM development implies the choice of software agent architecture and the specification of their behaviours. This development needs discussions between the computer scientist and the domain expert.


5.1. General specification process

The specification process is based on a new agent differentiation at the operational modelling level. The differentiation permits to define the simulation Multi-Agent System software architecture.

This differentiation is based on the two types of activities realized by conceptual agents: decision-making or operating activities. Decision-making or deliberative activities are based on decision models requiring informative interactions. Operating activities consist in executing an action. The decision-making activities and the operating activities of the conceptual agent are differentiated and affected to specific software agents. Each conceptual agent is associated to a generic architecture established from two software agent types.

The Cognitive-Agents perform decisional activities and communicate by messages exchange that corresponds to an information request or an order sending.

The Reactive-Agents perform operational activities and have behaviours defined by their states that specify actions they realized further to the reception of messages from Cognitive-Agents or signals emanating from the environment.

The operational agent modelling defines two agent societies in interaction, the Cognitive-Agent society or decisional society, and the Reactive-Agent society or reactive society. Each society groups together the agents of a same type. These different societies are to be implemented in specific software environments.

Each conceptual agent (Actor-Agent or Activity-Agent) generates a Cognitive-Agent in the OAM. To this Cognitive-Agent can be associated one or several Reactive-Agents. These represent each resource used for the object processing of the CAM. Links between the Cognitive-Agent and a Reactive-Agent are based on the notion of responsibility: a Cognitive-Agent is responsible of 0 or n Reactive-Agents (Labarthe et al., 2003). The exchanged messages between these agents correspond to the informational interactions and the signals to the physical interactions of the CAM.

To elaborate the CAM, a specific formalism has been defined in the previous section. For the OAM, a new formalism is proposed as illustrated in the figure 16.

Figure 16. Formalisms for the conceptual and the operational levels


The specification process is composed of the following modelling tasks:

1. Form a Cognitive-Agents society by representing all the conceptual agents (Actor-Agent and Activity-Agent) of the CAM by a Cognitive-Agent. Then, the informational flows between conceptual agents of the CAM are specified between Cognitive-Agents of the OAM.

2. Create a society of Reactive-Agents that represents the operative processes, represented in the CAM by objects. Then represent the interactions between Reactive-Agent and report objects that move on these interactions.

3. Define the responsibility links between the Cognitive-Agent and the Reactive-Agent.

4. Specify the objects that move on the Reactive-Agent interactions. 5. Specify the software agents, Cognitive and Reactive-Agents, with theirs

behaviours, according to specific diagrams. Each of these modelling tasks is detailed in the following sub sections.

5.1.1. From conceptual agents to a decisional society

To constitute the Cognitive-Agent society or decisional society, each conceptual agent (Actor-Agent and Activity-Agent) of the CAM generates a Cognitive-Agent in the OAM. The informational flows between conceptual agents of the CAM are specified between Cognitive-Agents. As detailed in figure 17, the Cognitive-Agents A and B correspond to the Actor-Agents A and B while the Cognitive-Agents ai and bj correspond to the Activity-Agents ai and bj.

Figure 17. Formation of the decisional society of the OAM


5.1.2. From conceptual agents to reactive society

The Reactive-Agent society, or reactive society, takes into account the physical interactions defined between Activity-Agents of the CAM. The physical resources of the CAM generate Reactive-Agents in the OAM. These Reactive-Agents are in interaction according precedence relationships in the processing process of the products, represented by objects in the CAM. These objects moving on the physical flows are represented by signals between Reactive-Agents.

As illustrated in figure 18, the Activity-Agents of the CAM generate one or several Reactive-Agents in the OAM. Reactive-Agents issued from these Activity-Agents allow the Cognitive-Agents to perform their activities. The interactions between these Reactive-Agents, modeled as signals, are specified with objects that move on the physical flows on the CAM. Note that the Reactive-Agents a11 and a12 are associated to the Cognitive-Agent a1, and the Reactive-Agents a41 and a42 are associated to the Cognitive-Agent a4. Two new intermediate objects, u1 and x1, have to be introduced for the interaction between these Reactive-Agents.

Figure 18. Formation of the reactive society of the OAM

5.1.3. Definition of responsibility links

The responsibility links between Cognitive-Agents and Reactive-Agents define the interactions between the decisional society and reactive society. In the OAM shown in figure 19, the Cognitive-Agent ai and bj are linked to Reactive-Agents by responsibility links. These responsibility links are specified by a cluster that includes the Cognitive-Agent and the Reactive-Agent(s) under its responsibility. Note that Cognitive-Agents A and B perform only decisional activities (no activity involving physical flow processing). Consequently no Reactive-Agent is attached by a responsibility link to these Cognitive-Agents.


Figure 19. Interaction between decisional and reactive societies

5.1.4. Specification of agent behaviours

The software architecture issued from the conceptual modelling will then be enriched by software agents more directly associated to the execution of the simulation. We have now to specify the knowledge, the behaviour of these software agents, and their interactions. In this software specification, various diagrams already used in agent oriented software engineering are exploited. The specification of agent behaviours requires specific formalisms depending on the nature of the agent.

As already used by several MAS design methodologies, some Unified Modeling Language (UML) diagrams, or extension of these diagrams to the agent context such as Agent UML (Odell et al., 2001), can be used to specify agent behaviours. Other formalisms can be also used, such as the formalism defined in the LSIS laboratory, Agent Behaviour Representation (ABR) formalism, (Tranvouez et al., 2001).

The behaviour of Reactive-Agents can be described using state charts. Reactive-Agent architecture is based on agent models of reflex type with states. Agent behaviours are implemented with state chart diagrams defined using UML / AUML to represent simple behaviour specified by a series of defined rules.

The behaviour of Cognitive-Agents is described with Behavioural Plans according to two models: individual and social. These plans are expressed in the form of state chart graphs or according to the formalism ABR. The individual models represent the autarkic behaviour of the agent (Local Plans), i.e. interactions with the other agents who are not required to reach their purposes. A social model involves the social behaviour of an agent (Role Protocols) and thus specifies how an agent interacts with others (message passing) as well as the actions leading to and resulting from these interactions. The set of Behaviour plans of an agent defines its ability to react and solve problems.


5.2. Operational agent modelling of the Golf case study

The OAM is composed of two agents' societies, one cognitive and one reactive. This decomposition allows differentiating the behaviour, the roles and the competences adopted by agents. Figure 20 presents the OAM based on the CAM (cf. Figure 15) for the conceptual agent «Agent Customized Golf Club» and the conceptual agent «Agent Finished Product Distributor». These conceptual agents generate in the OAM nine Cognitive-Agents and nine Reactive-Agents. The Cognitive-Agent, named «Agent Supply Golf Club», performs decision-making activities, involving complex behaviours. The Reactive-Agent, named «Agent Supply Golf Manage» performs operational activities, involving physical activities.

Figure 20. Operational Agent Model of the of the Fulfiller-Distributor interactions

The Agent Supply Golf Club has a Distributor role and is responsible for purchasing, receiving, distributing and managing inventory of golf club. These activities are: i) receiving and stocking golf club, ii) managing golf club purchases (ex: using a reorder method), iii) controlling inventory of golf club, and iv) distributing golf club by destocking. The interaction links between agents (physical and informative) are represented by objects (golf club) that move along the physical flows.

This agent has a behavioural plan to establish and to manage communication links with an agent responsible for golf club transport. These communication links are particularly achieved by sending order propositions, order confirmation or order refusal according to the received counterproposals.

Agent Customized Golf Club (Fulfiller )(Behaviour : Manage supplies and the Personalization Process)

Agent Finished Product (Distributor )(Behaviours : Manage Inventories and the Distribution Center )

Agent Customize Golf Club ( Fulfiller )(Behaviour : Schedule personalization process)

Agent Golf Club Stock ( Distributor )(Behaviours : Stock and Send Golf Club)

Agent Supply Golf Club ( Distributor )(Behaviours : Manage products and stock)

Agent Package Golf Club ( Distributor )(Behaviour : Schedule)

Agent Deliver Golf Club ( Distributor )(Behaviour : Plan Transport)

Agent Golf Club Manage ( Distributor )(Behaviours : Receive, stock and send Golf Club)

Agent Transfer Golf Club ( Distributor )(Behaviour : Prepare orders)

Agent Deliver Golf Club ( Distributor )(Behaviours : Plan Transport)

Agent Supply Golf Club ( Distributor )(Behaviours : Manage finished product and stock)

Agent Golf Club Manage ( Distributor )(Behaviour : Receive Golf Club)

Agent Personalized Golf Club Package ( Distributor )(Behaviours : Receive, Pack and Send Personalized Golf Club)

Agent Personalized Golf Club Manage ( Distributor )(Behaviours : Receive and send Personalized Golf Club )

Agent Personalized Golf Club Transport ( Distributor )(Behaviour : Deliver Personalized Golf Club)

Agent Golf Club Prepare ( Distributor )(Behaviours : Receive, Prepare and Send Golf

Club)

Agent Golf Club Transport ( Distributor )(Behaviour : Deliver Golf Club)

Agent Personalize Golf Club ( Fulfiller )(Behaviour : Transform Golf Club)


5.3. Implementation process

The implementation process permits to implement the MAS specified in the OAM. This system has to be executable in an agent software environment. This paper, centred on the methodological aspects, does not present in great detail the implementation process currently in progress.

Two different software environments are currently used to support implementation. The Cognitive-Agents are implemented on the platform MAJORCA developed in our laboratory (Tranvouez et al., 2001). This platform, developed in Java, integrates a rule engine JESS and complies with the FIPA ACL standard. The Reactive-Agent are implemented in the Anylogic© software (discrete event simulation). Figure 21 presents the implementation of the agents from the OAM. The implementation consists of transforming the agents obtained at the operational process by integrating them within a computer environment.

Figure 21. From operational modelling to implementation

A number of service agents have to be added to the previously defined software agents: i) the White pages agent (Agent Name Server), ii) the Yellow pages agent (Register agent), and, iii) the Event Scheduler agent. The Agent Name Server lists the agents occurring in the MAS, by loading in memory their name and the address, insuring the exchange of messages. The Register Agent is a repertory agent to whom each agent reports its name and skills. The Event Scheduler agent insures the synchronization of events and of the time among the agents of the OAM.

6. Conclusions and perspectives

In a context of stiff competition and high market and technological turbulence, with an emphasis on ever delighting customers, to maintain a competitive position, enterprises have to be demand driven and customer centric. Enterprises have to take into account the dynamic and stochastic nature of the demand, stemming from fast evolving customer expectations, to develop and exploit agility to meet the goal of

Signal

Message

Message

Events

EventsMonitoring

Agent

Agent Operational Model

Deliberative Agent

Computational Agent

Reactive Agent

MAJORCABehavioural Plans Representation

Communication

Reactive Agent Operative Behaviour Representation

Communication Agent

Anylogic

Agent

Decisional Society

Reactive Society


delighting their customers while maintaining profitability. Designing and managing supply chains in such a context requires the development of simulation-enabled decision support tools. The agent approach appears well suited for modelling such dynamic chains and for developing the required decision support tools.

This paper has presented an agent-based modelling for simulation of supply chains in a mass customization context. This approach allows a satisfactory representation of the structure and behaviour of the dynamic supply chain, specifically taking into account the operating policies induced by the stochastic demand patterns. Agents adapt their behaviours to the environment modifications. Hence, they are able to play different roles depending on the various management policies. The choice of the adapted role is guided by the decoupling point position in the supply chain. It allows the agents to determine the appropriate policy to ensure appropriate customer relationship according to the personalization level.

The proposed modelling approach is structured according to three modelling levels: Domain Modelling, Conceptual Agent Modelling and Operational Agent Modelling. According to these levels, three main methodological steps have been defined and presented in details. For each step we have introduced models and formalisms and various modelling processes. The application of the modelling approach has been illustrated on a case study from the Golf industry.

Currently we are developing a software platform which permits the implementation of the Operational Agent Model. This platform integrates two main software environments: the MAJORCA multi-agent platform and Anylogic©. The cognitive agents are implemented on the platform MAJORCA and the reactive agents are implemented on the platform Anylogic©. The exploitation of this prototype is a major enabling factor in a major research initiative on the design and management of mass customization supply chains. It particularly enables the rigorous study of coordination modes between actors along the chain, as well as the study of the operating strategies put in place to cope with the highly stochastic, dynamic and personalized demand.

7. References

Biswas S., Narahari Y., “Object oriented modeling and decision support for supply chains”, European Journal of Operational Research, vol. 153, no. 3, 2004, p. 704-726.

Booch G., Rumbaugh J., Jacobson I., Unified Modeling Language, Addison-Wesley, 1999. Childerhouse P., Towill D.R., “Engineering Supply Chain to Match Customer Requirements”,

Journal of Logistics Information Management, vol. 13 no. 6, 2000, p. 337-346. Brueckner S., Baumgaertel H., Parunak H.V.D., Vanderbok R., Wilke J., “Agent Models of

Supply Network Dynamics: Analysis, Design, and Operation”, in Harrison T.P., Lee, H.L., Neale J.J., (Eds), The Practice of Supply Chain Management: Where Theory and Application Converge, Springer Verlag, 2005.

Bussmann S., Jennings N.R., Wooldridge M., Multiagent Systems for Manufacturing Control, A Design Methodology, Springer-Verlag, 2004.


Disney S.M., Towill D.R., “A discrete transfer function model to determine the dynamic stability of a vendor managed inventory supply chain”, International Journal of Production Research, vol. 40 no. 1, 2002, p. 179-204.

Drogoul A., Vanbergue D., Meurisse T., “Multi-Agent Based Simulation: Where are the Agents?”, Proceedings of Multi-Agent Based Simulation, 2002.

Duray R., “Mass customization origins: mass or custom manufacturing”, International Journal of Operations and Production Management, vol. 22 no. 3, 2002, p. 314-328.

Fisher M., “What is the right supply chain for your product”, Harvard Business Review, March, 1997, p. 105-116.

Fishwick P.A., “A taxonomy for simulation modeling based on programming language principles”, IIE Transactions, vol. 30 no. 9, 1998, p. 811-820.

Frayret J.M., D'Amours S., Montreuil B., Cloutier L., “A Network Approach to Operate Agile Manufacturing Systems”, International Journal of Production Economics, vol. 74 nos. 1-3, 2001, p. 239-259.

Fox M.S., Barbuceanu M., Teigen R., “Agent-Oriented Supply-Chain Management”, International Journal of Flexible Manufacturing Systems, vol. 12, nos. 2-3, 2000, p. 165-188.

Goetschalckx M., “Strategic Planning Network”, in Stadler H., Kilger C., (Eds.), Supply chain management and advanced planning: Concepts, models, software, and case studies, New York, Springer, chapter 5, 2000, p. 79-95.

Gunasekaran A., “Supply chain management: Theory and applications”, European Journal of Operational Research, vol. 159 no. 2, 2004, p. 265-268.

Hoover W.E., Eloranta E., Holmstrom J., Huttunen K., Managing The Demand Supply Chain: Value Innovations for Customer Satisfaction, New-York, Wiley, 2001.

Labarthe O., Ferrarini A., Montreuil B., Espinasse B., “Cadre de coordination distribué de chaîne logistique par mesure des performances”, Congrès International de Génie Industriel, 2003.

Labarthe O., Tranvouez E., Ferrarini A., Espinasse B., Montreuil B., “A Heterogeneous Multi-Agent Modelling for Distributed Simulation of Supply Chains”, Proceedings of the International Conference on Holonic and Multi-agent Systems for Manufacturing, 2003.

Labarthe O., Montreuil B., Ferrarini A., Espinasse B., “Modélisation multi-agents pour la simulation de chaînes logistiques de type personnalisation de masse”, Conférence francophone de modélisation et de simulation, 2004.

Lakhal A., Martel A., Kettani O., Oral M., “On the optimization of supply chain networking decisions”, European Journal of Operational Research, vol. 129 no. 2, 2001, p. 259-270.

Min O., Zhou G., “Supply chain modelling: past, present and future”, Computers & Industrial Engineering, vol. 43, nos. 1-2, 2002, p. 231-249.

Montreuil B., Lefrançois P., “Organizing factories as responsibility networks”, Progress in Material Handling Research, 1996, p. 375-411.

Montreuil B., Frayret J.M., D’Amours S., “A Strategic Framework for Networked Manufacturing”, Computers in Industry, vol. 42, 2000, p. 299-317.

Montreuil B., Poulin M., “Demand and supply network design scope for personalized manufacturing”, Production Planning & Control, vol. 16 no. 5, 2005, p. 454-469.


Odell J., Parunak H.V.D., Bauer B., “Representing agent interaction protocols in UML”, in Ciancarini, P. & Wooldridge M., (Eds), Agent-Oriented Software Engineering, 2001.

Parunak H.V.D., Savit R., Riolo R.L., Clark S.J., DASCh: Dynamic Analysis of Supply Chains, Final Report, 1999, Center for Electronic Commerce.

Pine B.J., Mass Customization - The New Frontier in Business Competition, Boston, Harvard Business School Press, 1993.

Poulin M., Montreuil B., Martel A., “Implications of personalization offers on demand and supply network design: A case from the golf club industry”, European Journal of Operational Research, to appear, 2006.

Sadeh N.M., Hildum D.W., Kjenstad D., “Agent-based e-Supply Chain Decision Support”, Journal of Organizational Computing and Electronic Commerce, vol. 13 nos. 3-4, 2003, p. 225-241.

Tranvouez E., Ferrarini A., Espinasse B., “Multiagent modelling and simulation of workshop disruptions management by cooperative rescheduling strategies”, Proceedings of the European Simulation Symposium and Exibition, 2001.

van der Zee D.J., van der Vorst J.G.A.J., “A modeling framework for supply chain simulation: opportunities for improved decision making”, Decision Sciences, vol. 31 no. 1, 2005, p. 65-95.

Vernadat F., “UEML: towards a unified enterprise modelling language”, International Journal of Production Research, vol. 40 no. 17, 2002, p. 65-95.

Wooldridge M., Ciancarini P., “Agent-Oriented Software Engineering: The State of the Art”, in Ciancarini, P. & Wooldridge M., (Eds), Agent-Oriented Software Engineering, 2001.

Documents

A Methodological Approach for Agent Based Simulation of Mass …ACL 05.2... · 2018-02-18 · associated with modelling supply chains in a mass customization context. In section three,