[Zurück]

C. Gondowidjaja:
"Implementation and Coordination of Multi Agent Systems for Production Automation Simulation using Coordination Patterns";
Betreuer/in(nen): S. Biffl, T Moser; Institut für Softwaretechnik und Interaktive Systeme, 2008; Abschlussprüfung: 14.10.2008.

Production automation, e.g., for assembling more complex products from simpler parts, is a manufacturing process that coordinates a range of hardware entities (such as robots and transport systems) to achieve predictable system behaviour. Designers of production automation systems have many parameters to influence the properties of the overall production system, such as effectiveness and efficiency of the production automation system, which are hard to predict and costly to try out with real-world hardware. Multi-agent systems allow representing the hardware entities, their behaviour, and communication in a simulation environment, which facilitates efficient evaluation of design and parameter choices in order to optimize system performance.
Coordination describes the behaviour of a group of agents to optimize a known goal function while observing a set of constraints. A major challenge is to model the coordination of agent groups as agents may belong to several agent groups at one point in time or change group affiliation over time.

This thesis introduces coordination patterns that allow describing the coordination of agents on several levels in the context of a multi-agent simulation of a production automation workshop. The simulation consists of two major layers: 1. the business (or production planning) layer where the dispatcher transforms customer orders into more detailed work orders, which describe the machine functions in the workshop to carry out in order to fulfill a customer order, and prioritizes the work orders in order to optimize the overall system performance; and 2. the technical (or production automation) layer where the multitude of agents that represent hardware entities cooperate to carry out the current set of work orders coming in from the dispatcher.

Another aim of the development of the simulator and its coordination component is to test and analyze different workshop scheduling strategies which can me mapped to a real-world manufacturing plant.

Key contributions of the thesis are:
An architectural overview and guidelines for designers of multi-agent systems for tasks those are similar to production planning and automation. Coordination patterns describing the coordination of the production planning and automation layers in order to optimize the overall system performance. Design and implementation of a coordination component to simulate complex production processes and analyze the different production strategies. Design and implementation of a test case management system to run test scenarios with the simulator and measure the results for performance evaluation. A systematic evaluation of the performance of a range of production strategies and coordination patterns.

Major results of the work are:
1. The coordination with advanced production strategies is significantly more efficient than local optimization of agent behaviour; 2. The definition of guidelines for the development of a coordination component for agent-based production systems; 3. The development and implementation of a general interface, based on coordination pattern, between the business layer and the workshop layer for manufacturing system similar to the one used; and 4. Measurement of various workshop scheduling strategies based on the simulation results.