SAKARYA UNIVERSITY INDUSTRIAL ENGINEERING

Systems and Agent Systems Engineering Common Characteristic of Manufacturing Agent Automation from Point of Wiew PABADIS Project Instructor: Prof.Dr.Harun TAŞKIN By:Kerim GÖZTEPE 0650D06003 SAKARYA ;FALL 2006

1

Common Charakteristik of Manufacturing Agent Automation from Point of Wiew PABADIS Project Kerim Goztepe Sakarya University,Industrial Engineering . [email protected] FALL, 2006 1. Introduction The complexity of manufacturing and business environments,coupled with the complexity of tasks solved within these environments, are growing continuously. In many industrial scenarios, traditional centralized and hierarchical approaches applied to manufacturing control, production management, planning and scheduling, supply chain management and business solutions are not adequate and can fail due to their insufficient means for coping with the high degree of complexity and practical requirements for robustness, generality, and reconfigurability. These issues naturally lead to the development of new manufacturing and supply chain architectures and solutions based on the consideration of highly distributed, autonomous and efficiently cooperating units integrated by the plug-andoperate approach. This trend towards the application of multi-agent systems (MAS) techniques is becoming clearly visible at all levels of the manufacturing system and the businesses as a whole: On the lowest, namely the real-time machine control level, autonomous control units can now be found that are tightly linked or even fully physically integrated with the physical manufacturing hardware (e.g. with the conveyor, drilling machine, assembly cell) [1]. When solving management tasks (for example production planning and scheduling, strategic decision-making, data processing and so forth) on both the workshop and factory levels the application intelligent agents that draw upon the results of the artificial intelligent (AI) research can be used with significant advantage [2]. 1.1 Intelligent Agent An intelligent agent, equipped with a relevant amount of knowledge, can represent each manufacturing unit; the agents communicate, coordinate their activities and cooperate achieve an optimal, or at least plausible, solution in reasonable time (usually not done within hard real-time constraints). Generic visions of a deeper cooperation among enterprises, connected via communication networks, havelead to the ideas of virtual enterprises (VE). A VE is a temporary alliance of enterprises that come together to share skills or core competencies and resources in order to respond better to business opportunities, and whose cooperation is supported by computer networks [3]. The philosophical basis is the same for solving all of these highly distributed control and management problems: the community of autonomous, intelligent and goal-oriented units efficiently cooperating and coordinating their behavior in order to reach the global level goals (and maybe sometimes re-prioritizing their private goals to achieve better coherence, especially in the case of VE’s). In general and throughout this paper, an agent is defined as an autonomous, problem solving

2

computational entity capable of effective behavior in dynamic and open environments [4]. 1.2 Some Agent Architectures There are several typical agent architectures ranging from purely reactive agents (agents operate in a simple ‘‘stimulus-response’’ fashion) to deliberative or goaloriented agents, such as the ‘‘belief-desire-intention (BDI)’’ model of rational agents [5], which are proactively reasoning about their goals and actions. The terminology used in the field of agent technology is maturing but can be sometimes rather confusing. To reduce this confusion and provide better understanding, let us distinguish between: agents and agent-based systems in the sense that these are units/systems vaguely declaring properties of agency (namely they have the capability to act autonomously according to their reactive or proactive functionality and they have the capability to communicate among themselves) on one side; and MAS technology on the other side.

Figure 1:Agent platform

2. Multi-Agent Systems A MAS is a group of agents organized according to specific, precisely defined principles of community, organization and team operation. These principles encapsulate details of the agent architecture to be utilized, the style of inter-agent messaging to be adopted, the negotiation and cooperation protocols to be used, and so forth. This functionality must be supported by an adequate agent development and run-time platform that enables the agents to register/deregister within the community and allows them to offer their services and consume services from other agents in an open and flexible manner. Thus, an agent-based solution as mentioned in this paper should be viewed as a solution to a manufacturing control or business problem that considers the existence of agents as autonomous units. The use of MAS concepts and any MAS solution must satisfy the strict principles of agent community and organization, and so the application of principles in industrial environments represents a kind of technology. Intelligent agents are always considered as part of a MAS solution. Investigation of today’s state-of-the-art literature and discussions with the key practitioners in this domain have shown that there are three nearly isolated communities independently developing highly distributed, agent-based solutions in industrial environments: the community

3

designing MAS solutions for knowledge intensive decision-making on both the workshop and factory levels, that are endeavoring to meet the challenges associated with production management; and the community that is developing solutions for VE’s where a key challenge is to coordinate and integrate the existing manufacturing facility with external entities (sales networks, suppliers, customers, cooperators, and so on) to make better business level decisions. These three communities use different terminologies that in parallel focus on exploring different agent-based techniques. Because of the same nature of the background philosophy and the same principle ideas, there is a great chance (and hope) to link and cross-fertilize ideas between all three communities and apply their solutions, in an inclusive and integrated manner, to the problems identified on different levels. Namely it is a reasonable objective to use the same or compatible principles and techniques based on MAS principles across the real-time manufacturing level, the production control level, and the VE supply chain level. Let us illustrate that the MAS philosophy and technology provides a generic denominator and can help to pull together the isolated solutions from the three areas mentioned above. This integration vision leads to the idea of a crossintegrated enterprise (CIE), where the collaborative units are all the members of the control, manufacturing and supply chain considered horizontally (ranging from the control to business processes) across the entire (virtual) enterprise. This implies that manufacturing machines and cells, workshops, departments, enterprise divisions, suppliers and customers alike are agents within a MAS technology. In a CIE, the planning department of a supplier can directly ‘‘talk’’ to the controllers of a partner’s plant asking for actual data, or the workshop agents can directly negotiate with the workshops or even with controllers inside the outsourced company etc. Let us briefly summarize the results achieved by the three different communities in using agents and MAS technology to solve the particular industrial problems on their levels.

Figure 2: Multi-Agent System Architecture

2.1 The Management Level and Intelligent Agents On the production management level (which covers the tasks associated with distributed order pre-processing, production planning and scheduling processes, decision making on both the shop-floor and workshop levels, financial management, billing and so forth), the technology of intelligent agents developed in the area of distributed AI can be applied. 4

An intelligent agent represents each unit inside the production management level; namely, there is an agent to represent a workshop cell, one for a workshop, a storage unit, a production-planning department, a financial department, and so on. The agents communicate asynchronously in a peer-to-peer way. They solve mainly planning and scheduling problems, as well as decisionmaking in non-real-time higher-level control tasks. To solve these problems and make such decisions effectively, they are engaged in activities requiring deliberative behavior. An agent’s architecture usually consists of the agent’s body and the agent’s wrapper. The wrapper accounts for the inter-agent communication and any just-intime reactivity. The body is an agent’s autonomous decision-making component and is responsible for carrying out the internal functionality of the agent. In order to ease the interaction among agents, agents are often built using one of the common agent development platforms, for example JADE.This enables an agent to communicate in a standard way with the other agents in the system. The agent platform also provides the agent with facilities to understand an agent communication language (ACL) that expresses task requests and knowledge representation in a common format that the agent can understand and manipulate easily. The agentification process associated with installing agents into the production management level provides an elegant mechanism for system integration through using a tool to support the technology migration from the centralized systems towards the agent-based architectures. The communication among the agents is usually not just a random exchange of messages, but the message flow is managed by a set of standard negotiation protocols. These protocols range from a simple ‘‘questionreply’’, through the ‘‘subscribe-inform’’, to more complex negotiation protocols like the ‘‘contract net protocol (CNP)’’ or different versions of auctions (Dutch, sealed, Vickery etc.). Communication among the agents is an important enabler of their social behavior.The term social behavior means that the agents are able to: • communicate effectively with each other; • understand the common goals, states, and capabilities of the others; • honor the general rules and constraints upon behavior valid for each member of the community. These rules are called policies in Bradshaw et al. (2003) [6] or norms in Filipe and Cordeiro (2004). [7] 2.3 Agent-based scheduling in manufacturing systems Corresponding to the functionaland physical decomposition approach for agent encapsulation, two types of distributed manufacturing scheduling systems can be distinguished [8]: • those where scheduling is an incremental search process that can involve backtracking. Agents,responsible for scheduling orders, perform local incrementalsearche s for the orders they are in change and may consider multiple and multiskilled resources. The global schedule is obtained progressively through the merging of local schedule [9]. • systems in which an agent represents a single resource (e.g. a work cell, a machine, a tool, a fixture, a worker, etc.) and is responsible for scheduling this resource. This agent may negotiate with other agents how to carry out the overall scheduling [10].

5

At last, we can mention some production systems where the scheduling is completely distributed and organized locally at a product level. Under this condition, a local and simplified scheduling is performed by such an agent to accomplish a limited set of tasks with, some dedicated resources, long its production life cycle [11]. 2.4 Allocation through ‘‘protocol and bidding mechanism’’ Here, the allocation mechanism is generally realized through negotiations among agents using the ‘‘contract-net protocol’’. This protocol is a well-known example of such mechanism introduced by Smith (1980). It consists of five steps: a requested task is announced by a manager (from a given set of manager), potentialcontrac tors evaluate task announcements coming from several managers, potentialcontrac tors bid upon the selected task, the task manager awards the contract to one of the bidding contractors, finally the manager and a contractor communicate to execute the contracted task. This protocolcan be applied to the production system context in several ways: • managers and contractors are both resources, as for client-server. They can use subcontracting to distribute and dispatch the workload, • managers are ‘‘product offering work’’, and contractors are ‘‘resources bidding’’ to get the work assigned, • managers are ‘‘resources offering capacity’’ and contractors can be ‘‘parts bidding to use this capacity’’.

2.3 Limatations of Manufacturing Execution Systems (MES) Not only the field control is centralised in a common control structure, but also the manufacturing execution and the data handling. Integrated manufacturing execution systems (MES) have been designed to manage production at the shop floor level, and they have proven their efficiency by significantly reducing the manufacturing cycle time and hence increasing productivity. However, like any optimised and centralised organisation, MES suffers from limitations: Complexity: As manufactured products become more and more elaborate and various, the complexity of resource allocation and scheduling grows in an exponential way. Also the required efficiency and resulting quality of allocation and scheduling algorithms grows. Hence, centralised MES systems will become a bottleneck within the production control. Flexibility: The increasing variety of products and the fluctuation of a customer-driven demand necessitate expensive and time-consuming reconfiguration. To ensure system efficiency not only the integration of new product variations to the system, but also the exclusion of no longer used products has to be considered. Robustness: The complexity of operations raises the risk of failure, which brings the entire system to a halt. 3.The PABADIS Project ,MES and ERP PABADIS is an European IST project: ‘‘Plant Automation BAsed on DIstributed Systems’’). It is aimed at improving the management and control of a distributed and complex manufacturing system. The PABADIS approach focuses on automation using distributed systems. In fact, the aim is to dissolve the MES layer and divide its functionality into a centralised

6

part that can be attached to the ERP system and a decentralised part that can be implemented by partially mobile software agents. This effectively comes down to reducing the hierarchy to two layers. From the communications point of view, this approach both necessitates and supports the current trend in industrial automation to flatten the network hierarchy and to use IPbased networks down to the control level [12]. Briefly, PABADIS aims at creating a plug-and-participate environment which allows producing companies · to simply plug in a new machine and use it without major changes within the legacy systems and · to make job control more flexible by augmenting “conventional” (mainstream) ERP functionality with intelligence inherent in software agents.The baseline vision of the project is that every work piece has an agent “attached to it” carrying the necessary product information and moving through the plant the same way the work piece does. In fact, every production system needs two main ingredients: the actual physical work piece and information. If we consider a single piece production system, most of this information is tightly connected to the individual product, such as • production sequence and schedule, • machine-related production data, • status of the processing, • general administrative information about the order. In addition, there is information associated with the entire production system, such as • overall resource use, • overall production schedule, • machine status information, • quality control information. The system-wide scheduling and resource planning data should of course be consistent with the product-specific data sets, hence they can be compiled or deduced from each other. Traditionally, these data are generated by the ERP system in a strictly centralised fashion. Detailed planning and adjustments to the overall scheduling are subsequently done by the MES. With a view to the information distribution sketched above, it seems reasonable to largely remove the planning functionality from the ERP and distribute it on the level below among the “products” that can independently keep track of their processing needs and status. This requires the introduction of an information-oriented “alter ego” for each product, and software agents are a suitable approach for it. 3.2 Why using agents? PABADIS uses object-oriented models and object-oriented software technology to describe and perform automation tasks. The work piece is seen as an object that has all necessary information regarding its production somehow embedded or attached. It seems natural to use an intelligent Software Agent for such a purpose [13]. Software agents are the real-world manifestation of object-oriented and distributed functionality [14]. The combination of software agents and Deliverable Revolutionising Plant Automation: The PABADIS Approach IST – 1999 – 60016 © The PABADIS consortium 8 physical instances (like machines or the work piece in our case) is sometimes also referred to as “holon” [15], however, we prefer to stay with the term “agent” since in contrast to the machines we intend to put intelligence 7

also into the product by appropriate agents.Using software agent technology helps to design the system in a natural way that is easy to comprehend. Agents can be assigned to the physical instance they are responsible for. Other agents can represent and manage machinery or resources [16]. These agents inhabit a multiagent system (MAS) and can co-operate to perform their tasks. Co-operation and other methods from distributed artificial intelligence are further advantages that MASs can incorporate. Finally, the option to have mobile code adds another degree of freedom and improves the flexibility of the system. Other approaches like the classical client/server architecture can in principle also be used to implement distributed systems, but do not have the flexibility of agents.

3.3 PABADIS System Structure A PABADIS plant consists of fundamental building blocks, the Agency [PABADIS4], which is the PABADIS interface to an ERP system [PABADIS5], a set of Co-operative Manufacturing Units (CMU) [PABADIS3], which are used for product and data processing, a software agent community consisting of mobile Product Agents (RA) [PABADIS2], which follow the physical product and ensure there processing in co-operation with the CMU, stationary Plant Management Agents (PMA) [PABADIS2, PABADIS4], which ensure the product independent MES functions, and last but not least a SCADA CMU [PABADIS4], which is dedicated to SCADA functions. All building blocks are connected via a communication system and a Lookup Service, which ensures the system co-operation by providing contact capabilities [PABADIS6].

Figure 3:PABADIS Structure

8

3.4 Agent Types in PABADIS Based on such agent-oriented design, a PABADIS system basically consists of a set of so-called CMUs (Co-operative Manufacturing Units) that provide meaningful functions to the manufacturing process. This definition is deliberately abstract and makes no assumptions about the physical realisation. In fact, we can distinguish two different types of CMUs: Manufacturing CMUs are used for the physical processing of the products and involve some sort of manipulation of the work pieces. Depending on the granularity of the process steps and the envisaged level of abstraction, these entities can be individual machines (like dedicated drilling or milling machines), multipurpose manufacturing cells or full production lines. Even manual workplaces can be included in the system, provided they have a suitable HMI to the communication system. A special kind of manufacturing CMUs are transportation CMUs which link the other manufacturing CMUs and move the work pieces. Depending on the complexity of the plant, there may be one or several independent transportation CMUs. Logical CMUs provide computational services and have nothing to do with the physical processing of the products. Instead, they are consulted by the agents for special tasks like complex scheduling algorithms, database search, or the like. A special logical CMU can also be used to integrate SCADA functions into the PABADIS system.

9

REFERENCES [1] Deen, S. M. (Ed.). (2003). Agent based manufacturing: Advances in the holonic approach. Heidelberg: Springer. [2] Weiss, G. (Ed.). (1999). Multi-agent systems: Modern approach to distributed artificial intelligence. Cambridge, MA: MIT Press. [3] Camarinha-Matos, L. M., & Afsarmanesh, H. (Eds.). (2004). Collaborative networked organizations. Boston, MA: Kluwer Academic Publishers. [4] Luck, M., McBurney, P., & Preist, C. (2001). Agent technology: Enabling next generation computing. A roadmap for agent-based computing hhttp://www.agentlink.org/roadmapi. [5] Rao, A. S., & Georgeff, M. P. (1992). An abstract architecture for rational agents. In Proceedings of knowledge representation and reasoning KR&R-9 (pp. 439–449). [6] Bradshaw, J. M., et al. (2003). Making agents acceptable to people. In Handbook of intelligent information technology. Amsterdam: IOS Press. [7] Filipe, J., & Cordeiro, J. (2004). Organizational semiotics: A normative agentbased approach to VE modelling. In Processes and foundations for virtual organizations (pp. 271–278). Boston, MA: Kluwer Academic Publishers. [8] Shen, W., Norrie, D.H., 1999. Agent-based systems for intelligent manufacturing: a state-of-the-art survey. Knowledge and Information Systems, an InternationalJournal1 (2), 129–156. [9] Sycara, K.P., Roth, S.F., Sadeh, N., Fox, M.S., 1991. Resource allocation in distributed factory scheduling. IEEE Expert 6 (1), 29–40. [10] Shen, W., Norrie, D.H., 1998. A hybrid agent-oriented infrastructure for modeling manufacturing enterprises. In: Proceedings of KAW’98, Banff, Canada. [11] Diep, D., Massotte, P., Reaidy, J., Liu, Y.J., 2001. Design and integration of intelligents agents to implement sustainable production systems. EcoDesign 2001, 2nd International Symposium on Environmentally Conscious Design and Inverse Manufacturing, Tokyo, Japan. [12] D. Dietrich, T. Sauter, Evolution potentials fieldbus systems, Proceedings IEEE Workshop on Factory Communication Systems, Porto, 6.-8. Sep. 2000, pp. 43-350. [13] J.P. Müller, The design of intelligent agents: a layered aproach, Lecture Notes in Computer Science, Vol. 1177, Springer Verlag, Heidelberg, 1996 [14]J. Bredin, D. Kotz, D. Rus, Market-based resource control for mobile agents, Second International Conference on Autonomous Agents, Minneapolis, MN, May 1998, ACM Press, pp. 197-204.

10

[15] L. Bongaerts, J. Wyns, J. Detand, H. Van Brussel, P. Valckenaers, Schedule execution for a holonic shop floor control, in Proceedings of European Workshop on Agent- Oriented Systems in Manufacturing, Berlin, Germany, 1996 [16] M. Fletcher, S.M. Deen, Fault-tolerant holonic manufacturing systems, Concurrency and Computation Practice & Experience, vol.13, no.1, 2001, pp.43-70. [Gra] http://www.grasshopper.de

11