Sensitivity Analysis of the Thunder Combat Simulation Model to Command andControl Inputs Accomplished in a Parallel Environment

摘要

This research had two objectives. The first was to develop a methodology todemonstrate the parallel processing capability provided by Air Force's Aeronautical System's Command (ASC) Major Shared Resource Center (MSRC) and apply that methodology to the SIMAF Proof of Concept project. Secondly, AFSAA/SAAB requested a sensitivity analysis of THUNDER to the modeled command and control (C2) inputs. The power of parallelization can not be overemphasized. The data collection phase of this thesis was accomplished at the MSRC using a script developed to automate the processing of an experimental design, providing the analyst with a launch and leave capability. On average it took 45 minutes to process a single replication of THUNDER. For this thesis we made 1,560 runs in slightly less than 3 days. To accomplish the same number of runs on a single CPU machine would have taken slightly more than 3 months. For our sensitivity analysis we used a Plackett and Burman Resolution III screening design to identify which of 11 input variables had a statistical impact upon THUNDER. The decision to investigate only the significant variables reduced the number of input variables from 11 to 5. This reduced the number of design points necessary to obtain the same Resolution V information from 128 to 16 and eliminated the need for 3,360 THUNDER runs. A significant savings Using response surface methodology (RSM) techniques, we were then able to generate a response surface depicting the relationships between the input parameters and the output measures.