Knowledge-based Agile Assembly Architecture

Today`s manufacturing systems are often too inflexible and not sufficiently adaptable to rapidly changing environments with unpredictable and abrupt fluctuation in product demands or manufacturing downtimes. Moreover, such systems suffer from the weak covering and exchange of the information and knowledge between enterprise levels. Since the assembly participates with high percentage in the cost of manufacturing a product and since the automation rate in this domain is very low, this is where the most benefits can be gained by applying more flexible manufacturing paradigms. This project intends to develop innovative, agile control architecture to face the current requirements imposed to the manufacturing enterprises in the assembly domain. Having their own problem-solving capabilities and ability to interact in order to reach an overall goal, the autonomous agents are considered as a promising approach to provide a suitable paradigm for designing intelligent manufacturing systems to enhance flexibility and agility. We propose a knowledge-intensive multi-agent architecture that enables ontology-based communication and cooperation among a set of autonomous and heterogeneous agents. An agent, the main core of our architecture, acts based on his knowledge, by sensing the manufacturing environment, triggering the reasoning process, which selects the proper actions to be executed and that will affect the manufacturing environment. Each agent has knowledge about his domain of application, about strategies, which can be used to achieve a specific goal, and knowledge about the (other) agents involved in the system. The agent is a representation of a manufacturing component that can be either a physical resource (numerical control machine, robot, pallet, etc.) or a logic entity (order, supply, etc.). Having a shared ontology is critical for successful communication between agents, since such a shared ontology provides the common agreement and understanding about the concepts used. This offers the possibility for solving inter-operability problems. Therefore, the ontology will be shared among agents and will serve as the instrument to define the vocabulary used by the agents during their interactions, and to support understanding of the message content in the sense of its correct interpretation. In particular, the essential knowledge about the domain will be made available to the agents through an ontology.

The aim of this project was to investigate the applicability and effectiveness of a multi-agent system (MAS) approach for managing distributed manufacturing systems in the assembly domain. More specifically, the goal was to develop an innovative, agile multi-agent architecture, capable to handle ongoing changes in a rapidly changing system configuration and maintaining efficiency at the same time. The solution should support interoperability and reduce time and costs for getting high-quality and accurate knowledge through its information systems. It also has to be able to dynamically optimize a production schedule considering system capacity and time constraints. Moreover, the approach has to facilitate the easier incorporation of agents into industrial applications. The main results of the project can be split into several parts: the development of a knowledge-based multi-agent system as well as a single agent architecture, the design and employment of a persistent system ontology, as well as extensive simulation and real system tests to show the feasibility of our approach and prove its advantages. In order to reduce its complexity, we created four related agent types (contact, order, supply and machine agents) needed for proper functioning of the manufacturing system, addressing also responsibilities and activities of each agent type. Each agent in our architecture has an ontology-based world model, which role is to maintain the knowledge about the agent`s own activities in relation to its environment as well as to its underlying software parts. The ontology specifies the meaning of terms which are used during communication, enabling knowledge interoperations between agents. Taking into consideration the real world conditions, where particular actions have to be performed under real-time constraints, our architecture divides the control of machines - mechatronic components - in two parts: the underlying real-time capable low level control and machine agent - the more sophisticated, reasoning high level control. The feasibility of our approach was shown in a realistic industrial setting, the `Test bed for Distributed Holonic Control` at the Institute for Automation and Control, Vienna University of Technology. It served as a demonstrator for our distributed architecture. We also used simulation to analyze its impact on the system performance. We have studied the planning, scheduling, as well as operational abilities of our approach when applied in dynamically changing environments. The results of experiments in the simulation and on the real system underline increased reconfigurability, agility, robustness and fault tolerance. The applicability of our approach in dynamic, heterogeneous environments where accurate information and knowledge have to be shared among the system units is the most important outcome.