A study ln the use of distributed intelligent agents in the simulation of a flexible manufacturing environment

A relatively new AI technique is being investigated for use in distributed decision making. This method is known as intelligent agents. According to [Shohen 1993] a software agent is an entity that functions continuously and autonomously in a particular environment, often inhabited by other agents and processes. This behavior provides a framework in which to learn and adapt to changes in models. With this tool, we can add dynamic "thinking" into our models. For example, [Etzioni and Weld] say that distributed agents, once they have observed and perhaps learned something of their environment, will have attributes such as reactivity (the ability to selectively sense and act), autonomy (goal-directness and self starting behavior),, collaborative behavior (work in concert with other agents), adaptability (being able to learn and improve with experience) and mobility (being able to migrate in a self directed way from one host to another). Intelligent agents offer a unique method to control production flow and to make decisions in the flexible manufacturing environment (FME). The FME domain is a rich area in which to investigate the effectiveness of agents because the specific tasks of the agents can be well defined and the operation of the agents is well understood. In this paper we discuss using agents to route jobs through the FME. In particular, we focus on the coordination between the process agents (the individual jobs entering the shop floor) and the workstation agents (the resource needed by the job) in scheduling the route through the job floor and in balancing load throughout the job floor. The approach we take utilizes the notion of just-in-time (JIT) decision making. By JIT we mean that the process, in coordination with the workstations, dynamically chooses a "best case" next workstation to perform the next task. By using JIT, decisions can be delayed until necessary, allowing for more efficient decision making, especially under changing circumstances (i.e. machine failure or machine retooling). We have simulated our manufacturing agents using a testbed of PC and the programming language C++ (although a JAVA implementation is in progress). Preliminary experiments have been performed, and initial results are favorable when compared to other scheduling algorithms such as FCFS.