Optimization of multivariate simulation output models using a group screening method

Abstract Computer simulation is often used to identify the set of model variable values which appear to lead to optimal performance of the modeled system. When all variables are quantitative, experimental designs and procedures known as response surface methodology (RSM) can be used in dealing with this optimization problem. Previous research has emphasized the situation in which there are only two independent variables. If there are more than two independent variables, it becomes difficult to find optimization and the simulation processes may be expensive. Nevertheless, these are common circumstances. This paper proposes a methodology using a two-stage group screening experimental design to investigate which subset of the variables is most important in explaining the optimum response variable. A simulation of a new generation of flight simulator is analyzed to determine the values of six variables that will optimize the values of a dependent variable simultaneously. The group screening technique is a powerful and robust tool to implement response surface methodology strategies.