Dynamic output analysis for simulations of manufacturing environments

We describe the design and implementation of a generic, real time, in-line output analysis procedure for controlling simulations of discrete manufacturing environments. We implemented this capability in the commercial simulation software Extend(R). The main issues we faced were: (1) specifying the products to evaluate; (2) determining the batch sizes for output analysis; and (3) defining the stopping conditions based on the confidence intervals. We implemented a significance test for correlation and used this test to dynamically adjust the batch sizes used in confidence interval estimation done using batch means. When the stability conditions have been met, the simulation prompts the user to consider stopping the simulation. On the other hand, if at the end of the run length selected by the user, the statistical conditions were not satisfied, the tool notifies the user of that fact. This capability enabled us to significantly reduce the simulation run lengths, and ensures, with little additional computational effort, that the results were reliable. We used this tool to control simulations of electronics, steel, automotive, and metal processing industries. In general, using this tool we realized a reduction of more than 40% in the time required for simulation.