Agent-based information architecture for shop-floor control

Abstract

The principal aim of this paper is to develop a highly flexible shop-floor control system architecture using software agents which communicate with one another using the peer-to-peer paradigm. A shop-floor control system is responsible for the coordination of material and information flow in the manufacturing system as well as for communicating with the suppliers and the distributors or customers. It has to make several dynamic decisions like release of parts into the system, scheduling and routing of parts and transportation, machine selection, rerouting in case of failures etc. In our proposed control system such decisions are made by software agents which are equipped with local databases and independent computing power and a reliable communication interface. The different agencies and agents that make up the control system are discussed in this paper. In this proposed system, a part agent (as part of the part agency) enters the factory floor with process requirements, contacts the task assignment agent for the process plan, the monitoring agent for the status of the system, the scheduling agent for the detailed schedule, and the router agent for transportation before finally proceeding to get served. Mechanisms for handling issues like scheduling, routing, and failures have been discussed. Java and its underlying programming concepts form the backbone of our agent architecture. Agents use KQML (knowledge query and manipulative language) constructs for sending queries and KIF (knowledge interchange format) constructs for representing knowledge and exchanging information. Our system has several advantages over other control architectures and it is highly likely that future factory floor control systems will follow similar paradigms.