Collaborative Research: Procedures for Efficient Cycle Time-Throughput Curve Generation

Northwestern University, Evanston IL

Investigators

Abstract

This grant provides funding for the development of efficient and effective tools for the estimation of cycle time-throughput (CT-TH) curves via computer simulation. Unlike analytical queueing models, computer simulation can be used to represent almost any manufacturing system regardless of how complex. Unfortunately, simulation allows for the evaluation of only one point at a time on the CT-TH curve, so tools are needed to guide the simulation for development of the full curve. Since the CT-TH curve for real manufacturing systems is subject to variability, this research will identify and derive appropriate models and tools to represent not only the mean of the curve, but also its variance and selected percentiles, especially in the region of maximum system capacity. The tools for constructing the CT-TH curves will be adaptive and selective both in terms of the models used and the precision required by the user. Accurate model determination will be directed under both fixed-budget and fixed-precision settings. Experiments to evaluate the methods developed will be conducted on a test bed of tractable queueing models and a large-scale simulation of a semiconductor manufacturing facility. If successful, the results of this research will provide methods for efficiently generating the CT-TH curves associated with simulation models of semiconductor manufacturing facilities. Since CT-TH curves can be used to quantitatively evaluate different scenarios of product mix, production targets and capital expansion, the primary goal of this work will be to develop simulation methods for generating accurate CT-TH curves given finite resources. Compared to existing methods, this work will increase the number and complexity of manufacturing system evaluations that can be conducted leading to improvements in production facilities. Production improvements will lead to subsequent reductions in production cost and/or increases in throughput. The proposed work will also contribute to the tools available for model fitting, variance reduction and fixed sample allocation in large-scale simulation models.