ADAPTABLE REAL-TIME SCHEDULING USING ARTIFICIAL NEURAL NETWORKS

Over recent years there has been considerable work focused on enhancing the planning and control of manufacturing operations through 'real-time' reactive scheduling. A common characteristic of a large proportion of the developed systems is that they are only suitable for tightly prescribed control problems.To overcome this limitation a reactive scheduling system should have the ability to adapt its own computational process to the type of problems encountered and be applicable to a wide range of control situations. This paper discusses the on-going development of the architecture and underlying methodology used to build a real-time reactive production scheduling and control system, through the use of Artificial Neural Networks. The system, named ANNSR, has the capability to learn about the scheduling situation and can make accurate 'informed' judgements on 'new' conditions arising from the controlled environment. Furthermore, the ANNSR design means that ANNSR can be easily modified to handle a variety of the scheduling applications.