A neural network based control strategy for reconfigurable manufacturing systems

High-level production planning decisions are required for identifying basic courses of actions that form guidelines for control of manufacturing activities. For Reconfigurable Manufacturing Systems (RMSs), such decisions are complex since the system configuration is open and information about the current product of manufacture is often incomplete. In this work, the system configuration is cast as a virtual manufacturing structure consisting of processing stations whose task domain addresses a range of production scenarios and hence avail alternative process routings for parts. A strategy for identifying the combination of parts process routings that minimizes operating costs is outlined. Analytical functions for the strategy are developed through a combination of neural networks and the concept of similarity coefficients. Simulation experiments are conducted with search techniques that are employed to find the minimum entropy in a neural network architecture in order to obtain an optimal manufacturing schedule for flow of parts. The simulation study shows that the strategy is able to find optimum alternative routings for parts, from which the production volume matrix, process similarity coefficients and the processing required vectors are derived for use in production control. The results indicate that the strategy has potential in handling manufacturing activities in RMSs.

PDF (Full text) 38984.pdf - Published Version Restricted to Repository staff only Download (155kB)

Actions (login required)

View Item