Experimental design techniques for performance evaluation of generalized stochastic Petri net models

Generalized stochastic Petri nets have been used for the performance evaluation of manufacturing systems and other discrete event dynamic systems. The goal is to provide a framework for the designer of manufacturing systems where the designer can examine alternative design scenarios. The authors describe the application of robust design techniques to select only those experiments that have the most impact on system performance. Robust design techniques draw on many ideas from statistical experimental design. One such approach is commonly known as the Taguchi method. Here, the Taguchi method is used to efficiently determine the effect of a parameter on the system performance measure. An important by-product of this approach is that interactions between several parameters can be detected and the impact on the performance measure evaluated. This cross-effect is very important and is demonstrated on a simple transfer line and on a manufacturing system in which machine activities, database accesses, and network transactions are included in the model. The computational savings are also significant in these cases.<>