Tool for Rapid Analysis of Monte Carlo Simulations

Abstract

During the early design phases of any complex aerospace system, Monte Carlo simulations are performed and analyzed to better understand the physics of the problem and to identify design variables that must be either changed or studied further. The Tool for Rapid Analysis of Monte Carlo simulations was developed to assist engineers in the postprocessing of Monte Carlo simulation results. This work combines two pattern-recognition algorithms, kernel density estimation and nearest neighbors, into a practical analysis tool that ranks influential variables given a specific failure metrics. The Tool for Rapid Analysis of Monte Carlo simulations uses a failure metric to separate the simulation results into two groups: successful and failed simulation runs. The kernel density estimation and nearest neighbors algorithms are used to estimate probability density functions of the two different groups of data, which are then used to calculate a cost function that quantifies the relative influence of each design variable. In addition to producing a ranking of influential design variables, the Tool for Rapid Analysis of Monte Carlo simulations can rank combinations of up to four design variables in the form of differences and ratios. This paper shows results for a dynamical system with an analytical solution to demonstrate how these methods can identify failure regions in a Monte Carlo data set.