Huristic search methodology for compound simulation

• Authors: William J. Maddocks
• Venue: WSC '78 Proceedings of the 10th conference on Winter simulation - Volume 2
• Year: 1978

In recent simulation research related to complex automotive air pump assembly line operations, the major effort focused on finding the combination of sequential and parallel station capabilities that would minimize costs. A key variable is related to the in-process inventory that could reach very great proportions due to the high volume nature of the line. Within certain overall constraints, management can vary the output capability of each station by adjusting the number of work shifts scheduled. The in-process inventory also functions as a decoupling mechanism in this case. As no practical algorithm provides the solution, simulation of the line in GPSS V has been utilized. Here the essential independent variable is a decision vector (DV1...DVn) identifying the assigned shifts at each operating station. A simple stepping search FORTRAN subroutine is linked to the simulation model to provide the engine for changing decision vectors. In the context of a realistic line, the number of simulation runs can approach astronomical figures, consequently a rapidly converging search methodology is a real consideration. Preliminary experimentation shows promise of achieving effective convergence through the mechanism of reducing search step sizes and decision vector upper and lower bounds as the search process identifies improved vectors. A scaled down model i.e. 5 stations, is used for this study, however it retains the compound character of the original model. Selected decision rules are introduced and form the basis for the search methodology with the fundamental criteria relating to CPU time to achieve a target performance rating.