A Survey of Factory Control Algorithms which Can be Implemented in a Multi-Agent Heterarchy: Dispatching, Scheduling, and Pull

data types (queues, sets, lists, trees, ...) operations objects (windows, palettes, tool bars, ...) methods agents (radar stations, military equipment, factory machinery, ...) messages Table 1: The Progression to Agent Technology identical or complimentary agents which act together. There are a number of possible architectures which can be used to build multi-agent systems. 1.2. Multi-Agent Architectures Three types of architectures which are commonly studied in the multi-agent systems research community are functional, blackboard, and heterarchical architectures. Hierarchical architectures have also received a great deal of attention in the manufacturing systems research community. These different multi-agent architectures are described here with a specific focus on the heterarchical architecture. In a functional architecture each agent represents a functional capability. Usually there is only one agent per function. Functions communicate directly with each other, usually through main memory or as if through main memory over a network. Sometimes functions communicate on a peer-to-peer basis, but usually in a master/slave relationship where one function calls another function and expects specific results. Most commercial software is functionally decomposed according to structured analysis and design techniques [18]. Such software can be agent enabled by making each function an agent. Example functional multi-agent architectures which have been developed for manufacturing problems are provided in the references [19-22]. In a blackboard architecture, each agent has expertise in a certain area and the agents share their expertise by posting partial solutions to a problem on a central blackboard [23, 24]. The blackboard provides a persistent channel of communication, where past communications persist on Baker, Factory Control in Multi-Agent Heterarcies 4 May 8, 1998 the blackboard until modified by an agent. Example blackboard based multi-agent applications in manufacturing have included Maley’s CADENCE system [25-27], the OPIS system by Smith [28], and Boeing’s blackboard-based shop-floor control system [29]. In a hierarchical architecture, there are multiple levels of master/slave agent relationships where agents at one level of the hierarchy are slaves to a master agent at the next highest level of the hierarchy. In the Holonic manufacturing community, a hierarchy of agents (holons) is called a holarchy [30]. Early research in agent-based dispatching [31-33] and agent-based scheduling [3441] used hierarchical architectures as described later in this paper. In a heterarchical architecture [42-45] agents communicate as peers, there are no fixed master/slave relationships, each type of agent is usually replicated many times, and global information is eliminated. Heterarchical multi-agent architectures are the topic of this paper. Functional, blackboard, and hierarchical architectures differ from the heterarchical architecture in a number of important regards. Most functional architectures are not heterarchical because each functional agent is usually only replicated once, and functions often have fixed master relationships to slave functions. Blackboard architectures can not be considered heterarchical because of the use of a blackboard for global information storage. Though a heterarchical architecture may use global broadcast of information over a network, this information does not persist globally. Also, each agent in a blackboard architecture usually has a specific non-replicated function. Hierarchies differ from heterarchies in the use of fixed master/slave relationships. Also, some prominent hierarchical architectures [46] use global information repositories which violate the heterarchical goal of eliminating global information. Outside the manufacturing domain, a very active research field is to model as multi-agent heterarchies biological systems such as termite hills [47], ant colonies [48, 49], wasp nests [50], bird flocks [51, 52], fish schools [51, 52], and wolf packs [53]. Some inherent capabilities of these agent systems include self-configuration, scalability, fault-tolerance, and emergent behavior. That is, when termites are placed together, they self-configure to build termite hills. They scale up and down because if one adds more termites, they will join the hill-building process. These systems are fault-tolerant, if half the termites in the colony are destroyed, the remaining termites will continue building termite hills. These systems also exhibit what is called emergent behavior. That is, in simulations of termite colonies, it has been shown that individual termites can be identically programmed to follow a local behavior where when these termites are placed together they emergently perform a global result, a termite hill [47]. The complex behavior of building a termite hill emerges from the combination of termites into this heterarchical architecture, thus the term “emergent behavior.” Duffie claims other advantages of heterarchical architectures include reduced complexity, increased flexibility, and reduced costs [54]. The Holonic Manufacturing community views agent technology as allowing humans and manufacturing machinery to work together as colleagues. By Baker, Factory Control in Multi-Agent Heterarcies 5 May 8, 1998 putting agents on machines they can be made to work with people on a human level. Thus augmenting the performance of both man and machine and helping humans focus and leverage their talents on the aspects of manufacturing that they do best [55]. This is consistent with current trends in industry towards local empowerment and reduced levels of management [56]. The heterarchical architecture also enables massive parallelism. If agents are widely dispersed across a network of computers, then when these agents work together to solve a problem, they are bringing together these widely distributed computing resources to solve the problem. Finally, a heterarchical agent architecture maps best onto the Internet structure because the Internet is heterarchical. These features of multi-agent heterarchies are especially attractive to those looking to create agile manufacturing systems. Fully operational computer-integrated-manufacturing systems may some day be installed by merely plugging together manufacturing resources which are each managed and controlled by their own individual agents. The functionality of this agent architecture could supersede and integrate the functionality currently in the islands of automation in ERP, DCS, MES, SFC, PLC, and CNC software. These agents can be in a single factory or spread across factories over the Internet to form a single self-configured virtual enterprise. Though the advantages of emergent behavior, local empowerment, and machine operation on a human level may be claimed for functional, blackboard or hierarchical multi-agent systems; heterarchical agent architectures show a greater promise in terms of self-configuration, scalability, fault-tolerance, reduced complexity, increased flexibility, reduced costs, massive parallelism, Internet compatibility, and virtual enterprise formation. Though the potential advantages of reduced costs, reduced complexity, and better scheduling would benefit any manufacturer, the heterarchical agent approach could provide the greatest benefits to manufacturers who often need to change the configuration of their factories by adding or removing machines, workers, or product lines; to groups of manufacturers who at different times wish to form different virtual enterprises consisting of different subsets of their resources; to manufacturers who can not predict the possible manufacturing scenarios in which they will need to manufacture in the future; to manufacturers whose operations are significantly growing or shrinking; to manufacturers who wish to distribute their operations over wide area networks or who wish to participate in Internet-based electronic commerce; and to manufacturers who wish to empower their workers and provide them with the information and decision aides they need to make and implement distributed intelligent decisions. 1.3. Factory Control and Factory Control Algorithms If multi-agent heterarchies provide such a promising basis for future manufacturing systems, this naturally leads us to the question of whether we can control a factory using such an architecture. Baker, Factory Control in Multi-Agent Heterarcies 6 May 8, 1998