We describe a framework for analyzing simulation output in order to find solutions that will work well after implementation. We show how the use of a loss function that incorporates both system mean and system variability can be used to efficiently and effectively carry out system optimization and improvement efforts. For models whose behavior depends on quantitative factors, we illustrate how robust design can be accomplished by using simple experimental designs in conjunction with response-surface metamodels. The results can yield new insights into system behavior, and may lead to recommended system configurations that differ substantially from those selected by analysis solely on the basis of mean response.
展开▼
