Development of an intelligent shopfloor scheduler using neural networks and simulation

Abstract

In recent years, many firms have rediscovered the importance of scheduling in the shop

floor. Within the manufacturing functions, scheduling remains among the most

important and challenging tasks that must be performed routinely. Developing a

schedule involves designating the resources needed to execute each operation of the

process routing plan and assigning the times at which each operation in the routing will

start and finish execution. It is apparent that three common scheduling approaches:

OR-based, Simulation-based and Al-based alone cannot fully solve the scheduling

problem satisfactorily. The trend is towards a combination of the three approaches. A

hybrid approach using a neural network and simulation-based scheduling to solve the

detailed scheduling problem of the shop floor is presented. The neural network is

developed to analyze the complex information as well as orders coming into the

shopfloor and suggests candidate scheduling rules to the simulation model. The

simulation model then uses the rules to schedule the orders on hand. The scheduling

performance is analyzed, and strengths and limitations of the approach are discussed.

The work is set against a backdrop of a currently operating flexible manufacturing cell.