Simulation factor screening using cross-spectral methods (Q688218)

Detailed simulation models often contain a large number of input factors, many of which may be unimportant to the simulation output response. A preliminary procedure for determining which input factors may be unimportant is called input factor screening. This paper introduces a factor screening procedure which requires a single run of the simulation model regardless of the number of input factors. Whereas Frequency Domain Methodology (FDM) is based on an examination of the autocorrelation structure of the simulation output in the frequency domain, we consider a frequency domain analysis of the cross-correlation of the simulation input and the output using cross- spectral analysis. We develop a statistic using cross-spectral information to do qualitative sensitivity analysis in a designed experiment. In Section 2, we provide a modeling framework for cross-spectral analysis. The methodology is introduced and implemented for the linear effects of one input factor in Section 3. In Section 4, it is extended to linear effects of multiple input factors and the results are illustrated by an example.