7.1 Introduction/Motivation

Today’s markets are operating in turbulent and dynamic environments that are being influenced by permanent demands for higher-quality products and services at lower prices (Gunis et al., 2007). Growing product varieties and e-commerce opportunities have led to a paradigm shift from mass production to mass customization, which relates to the ability to provide customized products or services in high volumes and at reasonable prices (Anderson, 2004). Flexibility is needed at different levels, which may include the individual machine, the manufacturing system, the manufacturing operation (involving forming, cutting, or assembling), the manufacturing process, or the factory itself.

Commonly used technologies for shop floor applications do not offer this flexibility for manufacturing systems (Blecker and Friedrich, 2007). To cope with the lack of internal flexibility, manufacturing enterprises try to reach a high external agility and cooperate both vertically along a supply chain and horizontally among peers (even competitors) to form a multilayer, open, flexible, and cooperative production system (Hao et al., 2005). As a result, companies have to intensify their collaborative activities in order to maintain their position in the market (Schuh et al., 2008). To address the increasing complexity of collaborative industrial structures in highly dynamic environments, companies have to adapt the way they manage their operations (Colotla et al., 2003). Virtual enterprises (VEs) are one way that manufacturing companies are adapting their organizational and production paradigms to fit into this new collaborative production environment (Renna and Argoneto, 2010).

A VE (see Figure 7.1) is often defined as an integrated network of regular companies that join their core services and resources in order to respond to unexpected business opportunities and collaborate on an ad hoc basis. Such a network also includes suppliers, distributors, and retailers, which gather and share data and information about markets, customers, and internal competencies (Gunis et al., 2007). To allow the breakthrough of the VE concept, several research challenges—such as improved knowledge exchange and sharing, fast reactions to customer demand, reorganization capabilities, and the integration of heterogeneous entities—are required. Tools, techniques, and methodologies that will support interoperability, information search and selection, contract bidding and negotiation, process management and monitoring, etc., are also required (Camarinha-Matos, 2002). This means all the processes between enterprises, as well as within a company—for instance, from partner search and selection—including the negotiation between companies about a product through to the planning, scheduling, and production of it at the shop floor level, consider the actual status of the production line, scheduled orders, and ongoing business-to-business actions at any time (Koppensteiner, 2012). Due to the possible heterogeneity in a VE, it is usually not transparent to the partners which knowledge is available at which partner’s site. Even when availability is apparent, in most cases the knowledge is not understandable because of the different tools or formats employed. Stuckenschmidt and van Harmelen (2005) mentioned information search and representation as the two biggest challenges facing information technology. Moreover, the distributed nature of the VE sets requirements related to the supervision, coordination, and execution of local (company) goals, as well as global VE goals. The challenge is to introduce technologies that can support understanding as well as automation, and control processes connected with the creation, operation, and dissolution of VEs (Marik and McFarlane, 2005). Traditional technologies have difficulties in supporting automation in these complex VE environments.

The software agent approach offers a convenient way to deal with complexity, autonomy, and heterogeneity by providing the most natural means of representing distinct individuals and organizations, along with suitable abstractions. It has the ability to wrap legacy systems and offers the flexibility needed for organizational structure changes (Jennings, 2001). Hence, in the research community it is commonly acknowledged that autonomous software agents are a potentially fruitful way of approaching e-commerce automation. However, most multiagent systems (MASs) currently in use are designed with a single protocol explicitly hardcoded into the agents, and this results in a rigid environment, which accepts only agents designed for it. To overcome limitations based on simple syntactic matching, many approaches use ontologies to specify the representational vocabulary for a shared domain of discourse (Merdan et al., 2008; Tamma et al., 2002). This offers a way of representing concepts and relationships embedded in an environment such that they are semantically meaningful to a software agent. Moreover, the ontologies are used to describe the semantics of the information sources and make the contents explicit.

7.2 Characteristics of VEs

The main objective of a VE is to allow a number of organizations to rapidly develop a common working environment, thus managing a collection of resources provided by the participating organizations toward the attainment of some common goals (Martinez et al., 2001). This includes partner search and selection, VE contract bidding and negotiation, competence and resource management, task allocation, well-established distributed business process management practices, the monitoring and coordination of task execution according to contracts, performance assessment, and the interoperation of information integration protocols and mechanisms (Camarinha-Matos, 2002). The competitive advantage of the VE depends on the ability of the individual organizations to complement each other and their ability to integrate with one another (Meade et al., 1997).

Martinez et al. (2001) define three major characteristics that influence this kind of VE organization: market characteristics, production processes, and the strategic objectives of the association. All these characteristics are linked to each other. The market characteristics are influenced directly by customer demands or are the result of a market study for a new emerging market. According to these market characteristics, production processes will be set up, with a particular focus on coordination and information management. Some processes must be done in parallel as much as possible, because the time to market is the key parameter. Concurrent engineering and co-engineering can be used for this purpose. This leads to synergies between partners in the meaning of strategic objectives, such as common technology, commercial structure, production volumes, and resource sharing. The main concern about strategy could be cost reduction, reactivity (the ability to react to changes), or robustness (i.e., being able to satisfy customer demand no matter what happens).

7.3 The Benefits of Forming VEs

Internet developments, even collaborative paradigms, enable and introduce a lot of different types of VEs, which address various types of benefits (Camarinha-Matos, 2002; Hales and Barker, 2014):

• Agility: The ability to recognize, rapidly react to, and cope with unpredictable changes in the environment in order to achieve better responses to opportunities, shorter time to market, and higher quality with less investment.

• Strategy: By forming VEs, companies can concentrate on their areas of expertise and knowledge. This strategy releases the financial and human resources of companies, which in turn can be used to improve or reinforce the company’s strategic position in the market place.

• Finance: The formation of a VE allows companies to shorten their product development process and improve their time to market. Companies realize financial and human resource savings that may be used to satisfy other needs.

• Complementary roles: Enterprises seek complementarities (the creation of synergies) that allow them to participate in competitive business opportunities and new markets.

• Dimension: Especially in the case of small and medium enterprises (SMEs), being in partnerships with others allows them to achieve critical mass and appear in the market with a larger “visible” size. VEs benefit both small and large organizations. Small enterprises join forces and compete together for larger sections of the market that they otherwise could not access. Large enterprises, on the other hand, gain access to existing technologies and improve the flexibility of their operations.

• Competitiveness: Achieving cost effectiveness through the proper division of subtasks among cooperating organizations.

• Information and communication technology (ICT) infrastructure: A VE is built on interorganizational ICT systems that facilitate reduced transaction and switching costs, as well as data sharing and coordination across enterprise, geographic, and temporal boundaries.

• Flexibility: A VE’s flexibility means that it is ideal at redefining its scope and reconfiguring its resources quickly and concurrently to match market opportunities. This is based on its flexible linking to all existing and new partners and the quick delinking from departing partners.

• Organization: VEs operate in an almost flat and nonhierarchical structure. A flat hierarchical structure empowers the decision-making process at lower levels of the hierarchy, where problems are better understood.

• Resource optimization: Smaller organizations sharing infrastructures, knowledge, and business risks.

• Innovation: Being in a network opens up opportunities for the exchange and confrontation of ideas, which is the basis for innovation.

7.4 Obstacles and Approaches for VEs

ECOLEAD (Camarinha-Matos et al., 2008), as well as the international cooperation program intelligent manufacturing systems, is intended to establish the technological foundations needed to support VEs. Most of the research focuses on the vertical development of VEs, which targets collaborative design in manufacturing, dynamic supply chain management, service federation in tourism, etc. Only a few approaches target the horizontal development to establish the necessary base technology, tools, and mechanisms.

Some of the obstacles include the lack of appropriate support tools, namely for partner search and selection, VE contract bidding and negotiation, competencies and resources management, task allocation, well-established distributed business process management practices, the monitoring and coordination of task execution according to contracts, performance assessment, interoperation and information integration protocols and mechanisms, etc. (Camarinha-Matos, 2002). In fact, multiagent technology addresses issues, which fit into VE scenarios such as autonomy, interaction with others (as well as approaching inherently distributed problems with negotiation), and coordination capabilities. The nature of agents, by definition, enables a decentralized control of the enterprise. With regard to the VE life cycle, two different approaches of agents can be distinguished (Camarinha-Matos and Afsarmanesh, 2001):

• Agents in VE creation: a growing number of works are being published on the application of MASs and market-oriented negotiation mechanisms for the VE formation (Koppensteiner et al., 2011).

• Agents in VE operation: various projects have been addressing the dynamic scheduling and execution of distributed business processes (Koppensteiner et al., 2009; Rabelo et al., 2000).

Further problems include the lack of a common understanding of exchanged information among the cooperating organizations. Besides the long history of reducing communication problems through standards in the manufacturing area, the multifaceted nature of design information makes communication particularly different. In the meantime, ontologies are widely proven to enhance the information exchange in heterogeneous environments (Borgo et al., 2007; Merdan et al., 2008) and to share a common understanding in, as well as between, enterprises (Tamma et al., 2002).

The idea of VEs becomes of growing importance in manufacturing as an instrument to help companies face the challenges of quickly evolving market conditions, such as the following: very fast and continuous changes; new environment and working conditions regulations; improved standards for quality; fast technological mutations (Camarinha-Matos, 2002); changing customer demands for individuality and personal products. For the participation of manufacturing enterprises in dynamic cooperative networks, the capability to rapidly change the shop floor infrastructure is an important factor (Barata and Camarinha-Matos, 2003). Because of their un-agile structure, current control systems are a critical element in the necessary shop floor reengineering process, which implies the need for qualified programmers, something usually not available in manufacturing SMEs (Merdan et al., 2010).

7.5 MASs in the Coordination of VEs

As mentioned earlier, a VE represents a cluster of organizations collaborating to achieve one or more goals. MASs are very similar to the flexible network of enterprises and the entities that constitute a VE. In addition, the negotiation schemes used in the communication between agents in MASs can also be compared with the negotiation and communication between the member entities of a VE. It appears, therefore, that MASs may be particularly suited to the modeling of VEs (Pechoucek and Marik, 2008). Pechoucek and Marik see a long-term potential in the use of multiagent technologies in supply chain integration and the life cycle support of VEs.

The main challenge is to identify technologies that will help simulate, understand, automate, and control processes that are connected with the creation, operation, and dissolution of VEs (Marik and McFarlane, 2005). Several international research projects are investigating the deployment of agent technology in VEs, supply chain management, and interenterprise interoperability. In the CONOISE project (Norman et al., 2004), agent-based models and techniques are developed for the automated formation and maintenance of VEs. It aims to provide mechanisms to assure the effective operation of VEs in the face of disruptive and potentially malicious entities in dynamic, open, and competitive environments together with British Telecom. The European collaborative networked organizations leadership initiative (ECOLEAD) investigates and supports the deployment of technologies for the integration of collaborative networks of organizations, together with many industrial partners (Camarinha-Matos et al., 2005). Athena is another major EU-integrated project, contributing to the interoperability provisioning of the networked organizations and VEs, and the integration of a high number of industrial partners (Pechoucek and Marik, 2008).

7.6 MASs in E-commerce Applications

Intelligent software agents in e-commerce can be divided into the following: simple purchase or sale agents; complex buying and selling agents; and agent-based marketplaces. This classification includes an increase in the functionality of the agents. These simple agents function like information agents. A simple purchasing agent is concerned with information about the product and presents it to the user. This agent is also referred to as a shopping bot. In addition, complex purchasing agents have the function of ordering and paying for a product.

The highest level of agents in e-commerce is an agent-based marketplace that represents a further development of electronic marketplaces. An agent-based marketplace is a MAS in which sellers and buyers are represented by sales agents. In such electronic marketplaces, applications offer a wide range of functionalities for the electronic trading of products. These include user management, product catalogs, shopping carts, and secure online payment.

One of the best-known applications of agent-based auctions is eBay (Jennings and Rogers, 2009), which has had great success in recent years. Shopping agent research, for marketplaces such as eBay, dates to the web’s early years. In 1995 (Krulwich, 1996), which is often cited as the first shopping agent, let users compare prices of music CDs from Internet stores. PersonaLogic, another unbiased comparison-shopping agent, let users create preference profiles to describe their tastes, which allowed the shopping agent to identify products with features important to users (Menczer et al., 2002). eSnipe is an Internet business that automates a common bidding strategy on eBay called sniping13, based on software agents (Ockenfels and Roth, 2002; Kephart, 2002).

However, auctions are not useful for all types of markets and most of them usually offer only flexibility in prices. Other payment and delivery terms, quantity, and product characteristics, which are necessary within the cooperative production paradigm, are usually not negotiable. Automated integrative negotiation mechanisms have so far only been suitable for bilateral negotiations and have not yet been used for trading goods (Vetter, 2006). As a result, most of the developed systems have limited negotiation possibilities and auction capabilities. In these systems, prices can be changed by only one of the two involved parties, and complex negotiations involving more than two participants are not supported. Moreover, people are still involved in all stages of the buying process.

Although it is generally acknowledged that agent-based negotiation in e-commerce offers a step forward to automate negotiation tasks formerly done by human negotiators (Jennings and Rogers, 2009), procedures for business process automation are still required (Ko et al., 2009). The user should be assisted in carrying out time-consuming and complex negotiations. Agents can also facilitate negotiations between parties in different locations and time zones (Westkämper et al., 2007). Agent-based negotiations offer the ability to quickly adapt to the conditions of fluctuating market situations, to cope with the individual needs and preferences of business partners, and thus gain a competitive advantage (Whinston et al., 1997).

Vetter has presented a MAS based on an ontology that provides automated support for complex, multilateral negotiations in electronic markets (Vetter, 2006). The system allows improving the quality of service of e-commerce-applications and increasing the number of transactions. Experiences with simple agent-based marketplaces and the use of automated strategies for complex, bilateral negotiations in other areas of application make this agent-based approach promising for any kind of e-business solutions. To cope with the requirements in cooperative manufacturing, such as the scheduling and control of company internal processes related to ongoing negotiations, Vetter’s approach has to be enhanced with the abilities necessary for the manufacturing domain.

7.7 Case Study

The architecture presented in this section (see Figure 7.2) is based on Merdan’s layered architecture for manufacturing companies (KASA, Knowledge-based Multi-Agent System Architecture (Merdan, 2009)). To fulfill the requirements for flexible VEs, which need automated partner search and selection, as well as processes for the solution, operation, and dissolution of VEs, the system extends the management layer of KASA and adds two more layers to it: the VE creation layer and the VE operation layer.

• The VE creation layer handles the partner search and selection and is the first contact point for possible new companies. Therefore, this layer is responsible for communication between the different partners and the VE itself. Moreover, it handles requests such as incoming bids, requests for participating in an auction or B2B-negotiation, and requests for creating new auctions or B2B negotiations.

• The VE operation layer is responsible for the coordination of assigned auctions and links the market management of different companies with the VE’s operations. It defines different types of auctions, as well as their initialization and determination. Therefore, it handles the registration of all partners within the VE regarding auctions or negotiations.

• The management layer has been enhanced with the possibility for market management. Despite the normal understanding of the management layer, whose main focus is on internal production and resource initialization, as well as their determination, its communication abilities with the external environment are enhanced. This means managing the supply chain (with or without participation in auctions), as well as the participation and initialization of negotiations and VE creations while representing the company to other partners in the VE environment.

The structure presented defines the role of each layer in the system, as well as the associated tasks necessary to achieve common VE goals. This enables the creation of related agent classes and the mapping of these system goals to these agents.

Despite the trend toward decentralized systems, the architecture is designed as a central point in the control system (server) for two major reasons. (1) A central system providing the necessary knowledge to map different standards offers the possibility of easy integration with new partners (clients). (2) A centralized system enhances the data integrity, and therefore the safety, of all auctions. With a distributed auction system, it would be very difficult to ensure the integrity of all auctions, as well as the cautious handling of sensible information.

7.7.1 Multiagent VE Systems

Four types of agents are defined in the top three layers (see Figure 7.3). These agent types are optimized for the tasks they should fulfill within the different layers. The agents’ behaviors are based on rules, which are one part of the agents’ knowledge base. The other part is an ontology to ensure understanding between partners and to encapsulate the actual system status. Rules are simple if-then commands, and if all conditions for a particular rule are met, the rule will be triggered. Despite the fact that this concept looks very primitive, it is able to solve complex problems and can represent a large variety of negotiation strategies (Tamma et al., 2002). The rules themselves are hardcoded and predefined, but they can be easily replaced or extended; they offer great flexibility to the company’s negotiation systems.

The contact agent (CA) is a unique server-side agent that is responsible for communication between the different clients and the server itself. It registers new company-agents into the system and is responsible for the allocation of tasks to the system agent (SA), the information exchange between the server and the company-clients, and the handling of all incoming requests. Such requests can be incoming bids, requests for participating in an auction or B2B negotiation, and requests for creating new auctions or B2B negotiations.

The SA is another unique server-side agent and acts as a taskmaster. It is designed for all sorts of ontology-changing actions, such as the creation and destruction of auctions and B2B negotiations. The creation and destruction process also implies the management of the auctioneer agent (AA)—e.g., the creation and startup or the execution of the shutdown process (including the cleanup and removal of the AA from the ontology). It is also responsible for different server-specific requests, such as test or control messages.

AAs are server-side agents and are responsible for the coordination of their own assigned auction. The AA handles incoming requests from the CA, as well as communication with the market agents (MAs). Each AA is created in the same process as the auction and is also bound to it. It operates as long as it’s assigned an auction, and after it has started, it handles requests for its own auction.

Every MA is an interface between a company and the server. It represents the company and can be configured for different tasks, such as buying or selling goods within the VE. Either an authorized person or another agent can configure this agent. A MA can handle different tasks, such as auctions and negotiations, but it is also possible to configure more than one agent (e.g., one MA for each auction or negotiation). Once the agent has been initiated, it will start to work autonomously and, based on the configuration, it will either try to participate in an auction or start an auction by itself. It will place bids and react to events fully automatically; when the agent is initiated, no further input is needed. If a MA is supposed to participate in an auction, then it will analyze the current market situation and decide by itself which auction it should participate. The bid process then runs fully automatically.

7.7.2 The VE Ontology

For each company, it is important that it be able to define its own point of interest in a negotiation and estimate the reputation of other participants. In this heterogeneous environment of different companies from different countries, and possibly different laws, it is hard to capture all related concepts in a persistent ontology. Therefore, only the ontology part that is related to data exchange and communication processes should be isolated from internal company representations, such as goals, internal workflows, or points of interest. The presented ontology offers the representation and semantics of data about negotiation necessary to build VEs, as well as the handling of goods based on auctions, and presents a link to all other concepts in the company (see Figure 7.4). This negotiation ontology includes a description of basic concepts, such as process flows for orders, auctions, and negotiations, the users and tasks involved in this processes, and a description of products and services, as well as the interfaces used for internal company extensions.

The negotiation ontology has its roots in Vetter (2006) and acts as a general framework that defines the basic terminology, interaction, and protocols that enable agents to reach an agreement (Figure 7.5). Hence, it supports different negotiation types with multiple users at the same time in a VE. Besides the ability to support different auction types, auction properties, and negotiation tactics, it can also handle different products and services, their properties, and users.

An agent needs knowledge about all related concepts within the domain of the application in order to make adequate conclusions. To be able to reach its goals and accomplish the tasks with the best possible result, a decision about an auction or strategy has to be done according to all these attributes. Consequently, it is necessary to model and represent all these concepts, such as: the different possible auction types; negotiation processes and rules; and dependencies between agents and partners in the VE. Additionally, an agent has to know the user’s economic profile (the value of the product/service to be sold/bought, its total cost including shipment and production, quality, etc.) and its complexity and risk profile (size of the supply market, uniqueness of the product, availability, etc.). Agents are also externally influenced in terms of their relationships with other participants, which depend on each one considering the goals to be achieved or the tasks to be performed. Goals could be to buy at the lowest possible price or at the highest possible quality. Tasks could be participation in a negotiation or the supervision of an auction. Additional factors that influence negotiations with other agents could be the number of possible sellers/buyers, communication and information sharing, reputation, their reliability, and the quality or quantity of the product/service.

The ontology enables every user to have the possibility to start their own negotiation with an individually configurable MA, which can then handle multiple negotiations and related behaviors. These MAs then send offers to the AA. The AA takes these bids, compares them, and then sends messages back to all the MAs. These messages contain information about the state of the auction (the highest bid, remaining time, and so on). The MAs then evaluate whether they want to place a bid or leave the auction. When the time is up (or a maximum of negotiation steps is reached), the AA takes the winning bid (depending on the auction type) and creates a trade object. This trade object contains all the information about the seller, the buyer, and the auction itself.

7.8 Benefits and Assessment

In the VE case study being presented, the ability of the proposed architecture to manage the creation of collaborative enterprises, as well as the activities, is shown. The approach uses semantic technology together with software agents in order to improve knowledge capture, knowledge reuse, and knowledge transfer. The software agents are used within a complex virtual company to control certain components and processes (domains). Semantic means, in this context, that all relevant concepts that are important for collaborative organizations will be modeled in an ontology by capturing the associations between the domains, ensuring at the same time the understanding of the exchanged knowledge during the inter-agent communication. This allows business partners to build open communities that define and share the semantics of the information exchanged in their domain.

A multiagent architecture has been developed that is able to support the knowledge exchange and process control in collaborative enterprises. A commonly developed MAS for control of a manufacturing environment was extended with four additional agents, which were required to enable the automated creation and operation of VEs. The SA is required to enable the VE-system functionality and is responsible for the consistency of used ontologies. In this context, the CA is responsible for establishing a connection and managing the information exchange between companies. Additionally, the AA is necessary to start, coordinate, and end an auction. Besides that, each company is represented in this system by its own MA, which can be configured for different tasks, such as buying or selling goods for this company. This kind of MAS architecture has proven to be suitable for enabling the proper creation, coordination, and dissolution of VEs. The architecture being presented aims to help different companies fulfill their entire objectives by mediating the collaboration among the several organizations gathered into a VE.

In order to improve the automated data processing in a semantically heterogeneous collaborative environment, a persistent ontology has been presented, which is able to support knowledge exchange. The negotiation ontology, used by agents, acts as a general framework that defines the basic terminology, interaction, and protocols enabling the agents to understand each other and reach an agreement. This ontology-based approach enhances an agent’s flexibility in a dynamic negotiation environment and offers a simple and very comprehensive way to represent the reasoning capability of an agent. The ontology is also used to record actions and events as an explicit knowledge. As a result, this knowledge can be analyzed afterward.

Using this architecture with a semantic-enabled decision support system in an automated negotiation process offers a lot of advantages. The negotiation ontology allows access to an agent’s knowledge and analysis of its related data in order to support a suitable negotiation strategy. Using a history analysis and by slowly increasing its maximum offer after every lost auction, an agent is able to win negotiations on an average lower price than other negotiation agents. So the implementation of this kind of algorithm, even in a simple form, has shown advantages by achieving better auction results compared to systems that didn’t have such an approach (Koppensteiner et al., 2011). Besides, the negotiation agents have shown benefits due to their ability to consider more than just one attribute of a negotiation. Moreover, the developed semantic agent architecture enables related layers to communicate directly, avoiding unnecessary “stage by stage” procedures, because the agents from each layer are able to communicate and understand agents from any other layer (Merdan et al., 2011). The presented architecture assures clear definitions of each layer role in the system, as well as associated tasks that have to be done in order to achieve common goals. This further enables the smooth creation of related agent classes and the mapping of ultimate system goals to these agents.

The approach was also implemented in the mobile robot domain, which is also applied in a heterogeneous environment and depends on its accurate representation to undertake appropriate activity. The ontologies helped to structure a robot’s knowledge and its different levels of abstraction, as well as define concepts related to actions, actors, and goals. Additionally, the ontologies were used to test the knowledge sharing and exchange among the different types of robots in a real lab environment. The negotiation ontology was successfully tested to support mobile robots with negotiations and auctions concerning necessary tasks within a distributed disassembly of LEGO constructs. A low-level control and related interfaces for the MAS were implemented in order to support a real-time reaction capability (Koppensteiner, 2012).

7.9 Discussion

Bearing in mind the advantages of the multiagent architecture, it has to be noted that this technology has yet to mature through real industrial applications and thus establish a MAS ability to autonomously and faultlessly govern entire VEs. On the one side, the agents’ ability to maintain an accurate internal representation of pertinent information about the environment in which it operates has to be further developed. This could significantly improve its self-monitoring and self-control capabilities. In this regard, it is of vital importance to define the constraints that integrated subsystems (e.g., companies, clients, auctions, and control units) place on an agent’s world model representations, as well as to specify the means to measure the quality of ontological representation for autonomous agents. To ensure that the evolving semantic view of each agent is consistent with its mapped knowledge resources, there is a need to improve the evaluation mechanism of agents by incorporating the synchronization mechanism of the information on the existing status of a world model representation into the framework. On the other side, additional mechanisms need to be introduced that are able to capture expectations about future actions and states of environments, as well as monitor anomalies.

Moreover, automatic reasoning and interoperability in an open environment are often claimed as major advantages of semantic technologies. This approach provides the most flexibility because the agent can operate in an open environment and cope with unexpected cases. However, this requires a completely consistent, formal definition of the ontologies in use. In particular, various ontologies used within the system have to be properly aligned in order to provide the expected results. This is often impractical, mainly due to the complexity of the domains involved and the lack of experts in ontological modeling. Automated merging and mapping of ontologies can enhance the knowledge sharing and reuse. Considering the extremely distributed nature of a collaborative manufacturing environment, it is to be expected that multiple ontologies and schemes will be developed by independent entities, and coordination of those ontologies will require their merging and mapping. Ontology mapping involves mapping the structure and semantics describing objects in different repositories, whereas ontology merging integrates the initial taxonomies into a common schematic taxonomy. For this purpose, it is necessary to develop mechanisms that will be able to synchronize the changes of an ontology with the revisions to the applications and data sources that use their scalability to cope with the mismatches that may exist between separate ontologies. Additionally, it is of vital importance that these mechanisms support the extensibility and offer a diagnosis, or check the results of the alignments.

The proposed system was validated using a range of industrial-related case studies. The variation in products and solutions was considered in their selection. The analysis indicated that the agent-based approach and the ontological model of the VE are in line with the applications that are conducted in selecting and forming VEs. However, there is a need for flexibility in representing the types of agents that are applied in the VEs’ creation process. Although several improvements required for supporting VE formation were shown, the approach only can be used to represent a small percentage of generic attributes and automated matchmaking, and there is no general selection process. There is a strong need to be able to support a variety of processes and approaches, which can be defined by the user.

Future work will include the deployment of our current system into a real collaborative manufacturing framework, as well as its full exploitation on a broad spectrum of products and services. Regarding the flexible and dynamic nature of the VE, appropriate information technology and enabling services are required to support the establishment and management of the VE and the integration and interoperation of business processes. An effective information infrastructure is required to coordinate and enable the services in the VE life cycle in order to assist organizational decision makers across supply networks and accommodate the heterogeneity of data in terms of being able to connect to all forms of ERP systems and any other systems/databases.

To be able to ensure full integration of the presented approach in such complex environments, additional steps must be taken. Future work should strongly concentrate on the introduction of tools and techniques that will ensure the development of agent systems mature enough for industrial deployment. Moreover, risk management in a VE is an important issue due to its agility and the diversity of its members and distributed characteristics. Producers want partners able to offer products and services defect-free, reliable, and delivered just in time. A partner’s rating is needed to determine which suppliers are capable of coming satisfactorily close to this, and thus should be retained as current suppliers. The evaluation of partners within the VEs is crucial, where the partners that bid are sometimes VEs themselves and where the individual members within a coalition must be considered during the evaluation. In such situations, there is a need to be able to look into the coalition, as well as obtain a collective view of the coalition (Petersen, 2003). The procedure of a multifactor comparison—the combination of valuating the individual subjective criteria and the pricing and quality factors—must often be done manually, which is error prone and time intensive. Depending on the chosen level of complexity, usable tools for the supplier rating range from spreadsheets to cost-intensive and highly complex extensions of enterprise resource planning systems (e.g., SAP) are required. A solution is needed that is both supportive of the user throughout the automated steps and flexible enough to implement the different methods of supplier rating. At the same time, this should also be achievable for small and medium-sized companies.

In order to improve understanding of MAS emergent behavior, the focus is on the visualization of the interaction of MAS with the operator. The visualization will not only be useful for monitoring the behavior of the system, but also applicable to supporting the test process and the diagnosis functionality. If the diagnostic system detects a deviation from the common behavior pattern, the human operator’s expertise can be brought into the process to indicate the seriousness of the detected situation. This should also improve user trust in delegating tasks to autonomous agents. Moreover, the visualization of an appropriate interface can enable a link to a customer and support their engaging in the initial design of the products, as well as the manufacturing of these personalized products. The customer should be able to express and implement their requests and suggestions, as well as follow the production process (similar to how we follow shipments today). A further advance in the degree of automation of VE creation will increase the amount of communication and require a suitable security and privacy mechanism.

7.10 Conclusions

A distributed multiagent architecture, divided into different layers, is well-suited to support automation of the creation, operation, and dissolution of VEs. The distribution of layers in a company’s internal and external agents is important when it comes to user trust and acceptance of such systems. Especially for auctions, it is important to have an independent system that provides integrity and safety. The agents necessary for the creation and operation of VEs have to be separated in order to enhance the system’s functionality. To enable interoperability in heterogeneous VEs, it is necessary to provide a basic terminology. Especially in heterogeneous environments spread around the globe, it is hard to capture all necessary concepts in a single VE ontology. It is much simpler to isolate the part necessary for the formation, as well as the operation, of VEs in one commonly acknowledged ontology and to specify links to provided ontologies in companies participating in VEs (e.g., product descriptions).

To improve problem-solving abilities in multiagent architectures for VEs, decision support mechanisms should be provided. To support automated decisions during a negotiation process and to give an agent adequate suggestions, necessary attributes for such DSSs should be represented in an ontology. The MAS then has access to a given knowledge base and can use it in problem-solving processes. Moreover, the architecture has to provide defined links in the ontology, as well as in the agent’s behavior, in order to provide an easy integration of DSSs, as may be implemented by different participating companies.