Quick Detection of Influential Inputs in Monte Carlo Simulation

Abstract

Monte Carlo simulation (MCS) has been widely used in the development of aerospace vehicles. The rapid growth of computer power has allowed it to be applied in particular to the evaluation of systems before flight, where its capability of directly evaluating nonlinear systems incorporating various uncertain inputs is a key advantage. To improve a system’s design, it is crucial to detect those uncertain inputs that have a significant influence on unsatisfactory MCS results. This paper presents a new approach for detecting such input parameters quickly. It combines two statistical tests: Kuiper’s test and the $Z$ test. The advantage of the approach is that it uses only input and output data from MCS simulation evaluations: knowledge of the system model or further simulations are unnecessary. Because its results can be obtained quickly, the method contributes to the efficient development of flight vehicles. The approach also can be applied to MCS results that have a relatively small number of unsatisfactory results, so it is effective not only during the early stages of design where failures are more likely, but also for successive design iterations where failures are fewer.