An Artificial Neural Network as a Simulation Metamodel to Determine Production Parameters in a Specialized Manufacturing Setting

Abstract

The material handling system of a specialized production line is evaluated using a discrete-event simulation model. The object-oriented simulation model is built to allow changes to machine programmable logic, machine interaction effects, operator behaviours, and changeover policies. A second simulation model also allows for generalized production control parameters to be specified. The limitations of the current system are outlined and several opportunities for improving daily throughput are identified. To further understand and analyze the system, an artificial neural network simulation metamodel is developed and trained. To reduce the solution domain in training, a conditional Latin hypercube design is used. A strong network structure is identified through experimentation, and the performances of several training algorithms are compared. Finally, a simulated annealing algorithm is used with the trained and validated metamodel to determine reasonable production parameters for the system.