Gerstner Laboratory
for Intelligent Decision Making and Control 		 
 

PROPLANT
A Multi-Agent System for

Project-Driven Production Planning

Production planning is a complex industrial issue. We have tried to substitute classical planning and scheduling mechanisms with the process of negotiation, job delegation and task decomposition within a community of autonomous agents, each of which represents production or information unit of the modelled factory.

 

 

Project-Driven Production

We addressed the problem of Project-Driven Production where the most important emphasis is put on the design related activities. Instead of optimising the material and operational flow on a conveyor belt, the problem of project planning, project management, project design and configuration were the issue. Obviously, a manufacturing process of this nature is regarded as highly customer specific and an entire system depends mainly on pieces of software that shall be integrated. This is why we propose a multi-agent system as a principal methodology for Project-Driven Production modelling.

 

 

Multi-Agent System

Distributed Artificial Intelligence is a sub-field of artificial intelligence aimed at knowledge representation and reasoning process in a distributed environment. DAI has been rooted in diverse areas such as artificial intelligence, computer science, parallel computing, sociology, linguistics, or psychology. Multi-Agent Systems, one of distinctive sub-areas of distributed artificial intelligence, investigates behaviour of a collection of autonomous agents aiming for solving a given shared problem. It is very often the case that agents in multi-agent system are pre-existing, heterogeneous in nature, entities solving a particular subpart of the global task that would be otherwise out of capabilities of an individual agent (corresponds to high-qualified professionals).

 

 

ProPlanT - Production Planning Technology

We have proposed a prototype of ProPlanT - PROduction PLANning Technology that is aimed to utilising multi-agent systems in the area of project-driven production modelling. There are autonomous information processing units as building blocks of the production in a real enterprise. Behaviour of each of these is captured in a single agent in the multi-agent community. As an enterprise operation is hierarchically structured, there are three basic classes of agent considered (see Figure):

 

PRODUCTION AGENT (PA) - models behaviour of production departments, store administration, purchase department and other operating units within the factory. As a production agent we understand an encapsulated piece application (i.e. schedulling agents, database agents, etc.).

 

PRODUCTION MANAGEMENT AGENT (PMA) - models a managerial level, shared responsibility for the project outgrowth. These agents maintain an approximate model of the rest of the community and thus carry out optimal task decomposition and delegation. We have used tri-base acquaintance model for representing empathic knowledge of the agent.

 

PROJECT PLANNING AGENT (PPA) - carries out entire project planning related activities such as project configuration and design. This agent is a configuration expert system that posses a meta-represantion of the factory possibilities and configuration inference knowledge in a proof-planning like format.

 

META-AGENT (MA) - acts independently with respect to the community. It monitors agents behaviour and interactions and helps to improve the system efficiency by (1) runtime reconfiguration and (2) updating agent's' beliefs about the rest of the community.

See an overview movie of the ProPlanT system - overview.avi (9,298,454 bytes).

 

 

ProPlanT at Tesla TV

A prototype of the ProPlanT community of agents has been implemented for the purpose of modelling manufacturing process at Tesla TV, plc - TV transmitter manufacturer.

You can try and test a simple demonstration of the system at following link. The demonstration is fairly easy to set and run provided you have Windows NT operating system and TCP/IP protocol configured. Have fun.

 

 

Latest References

·         Pechoucek, M. - Lazansky, J. - Marik, V. - Stepankova, O.: Tri-Base Acquaintance Model for Project Driven Production Modelling. In: Changing the Ways We Work. Amsterdam: IOS Press. 1998. pp. 623-631. - ISBN 905199426-5

·         Pechoucek M.: Proof Planning Structures for Industrial Decision Making Formalisation. Submitted to PAKEM´99 – Practical Application of Knowledge Management, London: The Practical Application Company.1999

·         Marik V., Pechoucek M., Lazansky J., Roche C.: PVS’98 Agents: Structures, Models and Production Planning Application, In: International Journal of Robotics and Autonomous Systems special issue on Multi-Agent Systems Applications North-Holland, to appear, (available also as Technical Report of The Gerstner Lab CTU Prague & FAW Linz – Hagenberg – Praha - Wien, December 1997 - GL-76/97)

·         Marik V., Pechoucek M., Stepankova O., Lazansky J.: ProPlanT: Multi-Agent System for Production Planning, In: Applied Artificial Intelligence, to appear, (available also as Technical Report of The Gerstner Lab CTU Prague & FAW Linz - Hagenberg – Praha - Wien, December 1997 - GL-76/97)

·         Hazdra T., Pechoucek M.: ProPlanT – A Multi-Agent System for Operative Planning of Production. Submitted to PAAM´99 – The Practical Application of Intelligent Agents and Multi-Agent Technology, London: The Practical Application Company.1999