A Design of Experiments-Based Method for Point Selection in Approximating Output Distributions
McCormick, David Jeremy
Olds, John R.
MetadataShow full item record
The goal of this research is to find a computationally efficient and easy-to-use alternative to current approximation or direct Monte Carlo methods for robust design. More specifically, a technique is sought to use selected deterministic analyses to obtain probability distributions for analyses with large inherent uncertainties. Previous research by the authors has presented a promising class of methods known as Discrete Probability Matching Distributions (DPOMD). This paper introduces a new type of DPOMD better suited to problems with larger numbers of random variables. This new type utilizes a fractional factorial design of experiments array in combination with an inverse Hasofer-Lind standard normal space transform. The method defines points in the problem space that represent the moment characteristics of the input random variables. This new method is compared to two other approximation techniques, Descriptive Sampling and Response Surface/Monte Carlo Simulation, for three common aerospace analyses (Mass Properties and Sizing, Propulsion Analysis and Trajectory Simulation). A Monte Carlo analysis with corresponding error bands is used for reference. Preferences for probabilistic analysis each of these problems are determined based on the speed and accuracy of analysis. These results are presented here. The new DPOMD technique is shown to be advantageous in terms of speed and accuracy for two of the three problems tested.