Evaluation of a Neural Network Approach on FMS Cyclic Task Scheduling Problem

K.S. Low and M.-T. Kechadi (Ireland)


Planning and Scheduling, Neural Network, Cyclic Scheduling, Flexible Manufacturing Systems, Work In Progress Inventory


In this paper, we evaluate the application of a recurrent neu ral network (RNN) approach to the cyclic scheduling prob lem found in the Flexible Manufacturing Systems (FMS) environment. By studying the schedule of start times, the goal is to minimise the Work In Progress (WIP). The RNN approach is suitable to be used in cyclic scheduling prob lem of this type due to the NP-hard classification of it. Firstly this approach will describe details in calculating WIP from the cyclic schedule while integrating the corre sponding constraint cycle, linked, precedence and disjunc tive constraints found in the problem. A model based on an unconstrained optimisation type is formulated, followed by its solving methodology, which will be used to build its cor responding neural network. A corresponding perturbation algorithm is introduced to further improve the solutions. To validate this approach, three significant benchmarks are modelled, simulated and the experimental results analysed. A conclusion is finally discussed showing the effectiveness of the approach in this paper.

Important Links:

Go Back