A neural network approach for the real time control of a FMS

We propose a three phased model of flexible manufacturing system control. The first phase is to identify the feasibility of job moves under a given system status. The simple Sigma-Pi type of neural net model has been adopted in Phase I for feasibility recognition. The second phase applies the Hopfield-Tank model to determine the most appropriate job moves from a feasible job set derived from Phase I. This problem is considered to be very difficult not only because of its NP-complete feature, but also because of the need for quick response under a real-time environment. A Hopfield-Tank network with a linear energy function is proposed for Phase II. Phase III devotes to routing decisions for MHS. A heuristic algorithm based on Kohonen´s self-organizing feature maps is proposed.<>