Chapter 5. The new generation of manufacturing
5.7 Comparison between Multi-Agent Systems and Holonic
The research communities working in Holonic Manufacturing Systems and Multi-Agent Systems fields approach the problem of "intelligent manufacturing" from different viewpoints and nearly independently (agent technology was firstly implemented to realise heterarchically controlled systems, and later on, the concept of holonic systems was introduced to address more specific requirements for manufacturing systems). They use their specific terminology and techniques. Both the paradigms share some ideas and they differ in the other issues.
The debate on clarifying the difference between holons and agents is an ongoing issue in the research communities using these paradigms. Given the essentially different path on which each concept was developed the question itself is inappropriate [Ulieru, 2002] . In the following we briefly present the main similarities and differences between the holonic and multi-agent paradigms.
5.7.1 Similarities between Multi-Agent Systems and Holonic
Manufacturing Systems
The vision of a holonic factory draws a number of its concepts from the world of Multi-Agent Systems. That is why many similarities can be identified between these two areas.
Both the research communities do respect the same, very fundamental principles of holons' and agents' activities such as their autonomy, cooperativeness and openness. [Marik & Pechoucek,
2000] .
� Autonomy -An entity (agent!holon) has the ability to operate independently of the rest of the system and possesses some kind of control over its actions and internal state [Ulieru, 2002].
� Cooperation - Through the cooperation process entities (agents/holons) develop mutually acceptable plans and execute them. [Ulieru, 2002]
� Openness/reconfigurability -integration of new systems or remission of existing systems can be achieved without stopping the process.
Both approaches provide most of the characteristics listed in Section 5.3 "Main requirements of the new generation of manufacturing control systems" (modifiability, extendibility, adaptation, fault-tolerance, etc.) and both, holons or agents, have multi-layered architectures. [Marik & Pechoucek, 2000]
Goran D. Colak Chapter 05 There are similar trends in standardisation which are quite evident with the IEC (International E lectrotechnical Commission) 6 1499 standard in the case of holonic systems and the FIP A (Foundation for Intelligent Physical Agents) standard in the area of Multi-Agent Systems.
5.7.2 Differences between Multi-Agent Systems and Holonic
Manufacturing Systems
The holonic system community is rooted in the concept of holons as presented by Koestler and is strongly driven by the requirements of industrial control. The community is well organised around the international HMS (Holonic Manufacturing Systems) consortium. On the other hand, the comparatively much larger and more diverse community of researchers working in the Multi-Agent System (MAS) area is influenced by the ideas of highly distributed computing in computer networks as well as by the ideas of distributed artificial intelligence. As the community is much more heterogeneous, there are different organisational frameworks where the researchers are grouped. The European MAS researchers are organised in the AgentLink consortium, worldwide in IFMAS (International Foundation for MAS), Agent Society, and FIP A with an emphasis on industrial standards [Marik & Pechoucek, 2000] . Holonic and multi agent approaches differ in the following:
);;> Differences in a concept origin and emphasis: Holonics is an organisational paradigm inspired by the self-organising properties of natural systems. The emphasis is being on the structure of components rather then on the interaction between them. On the other side, agents have been envisioned as a software paradigm aiming to expand the limitations of the static object model with proactive capabilities of autonomy and environmental awareness. MAS aims to represent dynamical systems in software by focusing on the interactions between their parts - software components modelled as software agents - rather then on their structure [Ulieru, 2002].
);;> Nature of systems elements: In the manufacturing domain, holons have been defined as entities consisting of an information processing part and an optional physical processing part. On the other hand, a software agent is exclusively a software entity .
);;> Interest: The primary interest of HMS is in building a physical shop floor architecture, which is composed of co-ordinated and co-operating, interoperable and reusable hardware/software field components. MAS community has concentrated (until recently) on information agents only.
);;> Differences in a n ature of interactions among elements: In a holonic system cooperation is a precondition for the existence of the holarchy per se. Cooperative interactions among holons bind the holarchy together driving it towards the achieving of common goals with maximum efficiency. On the other side, in a MAS there is no pre assigned condition that the interactions among agents should be driven by cooperative forces. In a MAS agents may interact based on competitive rather than cooperative rules (such as electronic markets - competitive/conflicting environment) [Ulieru, 2002].
In addition to above differences, considering motivation, subject of research, usage of holarchy principle, and implementation of human interface, Marik and Pechoucek reported the following quite evident and distinguishable differences [Marik & Pechoucek, 2000] .
);;> Motivation: The holonic research is motivated by pragmatic manufacturing control requirements, on the opposite side, the agent research is motivated by implementation of distributed computational systems and decentralised decision-making.
);:> Subject of research: The holonic system researchers are preferably oriented towards the low-level end of the manufacturing process, low-level communication and behavioural standards, integration, etc. Multi-Agent System researchers aim at implementing social behaviour of intelligent entities, cooperation and coordination strategies, intelligent brokerage, learning from ones own experience, teamwork and coalition formation, etc. From a very simple viewpoint, we can see the holonic system research stream providing platforms/frameworks for implementation of knowledge driven higher level coordination and communication strategies based on the MAS research results.
);:> Usage of holarchy principle: The holarchy principle, which allows the creation of a holon as an integrated set of lower level holons, is used in HMS. This is not considered in the MAS field where autonomy and functional differences of individual agents are preferred. However, agents very often group themselves into hierarchically organised teams.
);:> Human interface: Each holon is usually equipped with a human interface. Human interfaces in MAS are very often implemented as separate agents providing services to the community as a whole.
5.7.3 Holonic Manufacturing Systems and Multi-Agent Systems
integration
There are no conventional methods for realising the holonic characteristics. To realise a holonic manufacturing system, comprising numerous interacting holons, special techniques for task executing and problem solving have to be applied. From a software engineering perspective a holon, as a unit of composition retaining the characteristic attributes of the whole system (holarchy), can be viewed as a class. Thus the object-oriented paradigm seems suitable for modelling holarchies as software systems. However, because of its characteristics (of autonomy, cooperation, and pro-activeness at the first place), an agent is nearly predestined for implementing holons. Thus, a Multi-Agent System appears an even more suitable tool for emulating holarchies as software systems than an object-oriented model. Through the concept of 'partial cloning' characteristics from a real physical entity (a holon) which are needed for pursuing collaborative actions in holarchy are abstracted and encapsulated in a software entity (an agent). Thus these software entities, which emulate the physical entities, enable the coordination of production through intelligent control procedures [Brennan, 2000] . A MAS which emulates a holonic system will consist of specialised autonomous agents (which have a particular structure and holonic properties), driven by a coordination mechanism designed according to the rules for cooperation of the respective holarchy. With this in mind it is easy to point out that software holarchies are specialised MAS 's that define the interaction between their agents based on the underlying cooperative holonic units [Ulieru, 2002] .
The issue is not "convergence" of the two paradigms but rather the implementation of an organisational holarchy (real life system) into software using the MAS paradigm. It should be noted that holonic manufacturing is not an alternative nor an identical approach to multi-agent control but rather it is complementary in that it represents a systems engineering approach to the development of manufacturing control systems infrastructure, rather than a solution mechanism for solving individual manufacturing control problems [Ulieru, 2002] .
Figure 5-5 describes a potential pattern of two holons, where the "intelligent" control module is implemented by agents [HMS, web]. The decision-making unit of the holon and the communication with other holons and humans, and optionally the physical entities are all adopted by agents. To take decisions an agent needs some information such as knowledge about
Goran D. Colak Chapter 05
the outer world or other agents. The state makes the agent react in a different way to similar inputs. It can be changed corresponding to some rules, which are based on knowledge. Obviously the MAS techniques are convenient for realising holonic systems. In fact, most developments of holonic systems to date have deployed agent-like solvers as a means of resolving planning, scheduling and shop floor control issues [Ulieru, 2002].
Decision making Communication
Human Physical Inter -
Interface control holon
Decision making Communication
Human
Interface
a) General architecture of a holon
State Knowledge Rules State Knowledge
Communication Communication
Human
Interface Physical Inter -
Human Physical control holon Interface control
Hardware
b) Agent-oriented architecture/or a holon Figure 5-5. A holon architecture
Rules
Inter - holon
5.8 Design principles for highly distributed
heterarchical Shop Floor Control systems
For designing highly distributed heterarchical systems Prabhu reported the following design principles: [Prabhu, 2000].
};> There should be no master scheduler. Complexity due to inhomogeneity of levels 1 would be reduced to give much simpler and more manageable problems. The benefits of
1 The control equipment that needs to be interfaced usually comes from many diverse sources - different brands.
reduced complexity would include reduction in development cost, and improvement in the chance of success.
)> Scheduling and control must be contained within the same logical entity. Such an approach would ensure the highest level of local autonomy. The most important information required for scheduling would be available within the entity, avoiding the need for using time critical messages. Also modularity of the system would ensure that the system is easy to modify and integrate.
)> Entities m ust follow the principle of least commitment. The establishment of relationships must be delayed as long as possible and terminated as soon as possible. Uncertainty problems caused, for example, by changes during production would be, therefore, be partly solved.
)> Entities must co-operate with other entities whenever possible. This is a rational requirement of a heterarchical system. Such an interaction is necessary to solve resource sharing conflicts in the best possible way.
)> Entities should abide by the schedule whenever possible. In the system as a whole, unnecessary surprises, which might occur if any entity pursued its own agenda, would be eliminated. This is in accordance with the spirit of co-operation in the system.
)> Entities should not assume that other entities will abide by their schedules. Loose coupling between entities should be ensured. In return, this would increase implicit fault tolerance.
)> Entities should not need to know the schedules of other entities. Global data should be avoided. Dependency on other entities would be minimised and implicit fault-tolerance would increase.
)> Entities must not make any assumptions about the system characteristics. (For example, infinite buffers, machine failure characteristics, processing times, etc.) Global information would be minimised and implicit fault-tolerance would be further increased. Also, this would increase the system flexibility.
)> Entities must be under minimum design constraints. There should be no predetermined routes or sequences between entities built into the entities themselves. Therefore, system flexibility could be used to cope with shop floor uncertainties.
)> Entities must autonomously decide trade-offs between local and global performance.
In this case the highest level of autonomy would be ensured and communication requirements would be minimised. For this process to occur, global information is required as discussed below.
)> Entities must form the local schedules with a global perception. To ensure system survival, instead of local, entity goals, local schedules must pursue overall system goals. Global perception of the system is the essence of co-operation in heterarchically controlled systems.
These principles are important considerations in the design of the system presented in this thesis (see Chapter 1 0). The next chapter, therefore, is dedicated to a brief overview of Multi-Agent Systems which have been seen as a promising approach for developing distributed manufacturing control system.
Goran D. Colak Chapter 06