Methodology for the validation of aircraft simulations in a mission scenario through the use of a validation tree with uncertainty

摘要

A methodology is developed that uses probabilistic analysis and computational simulation to continually update the state of a fleet of aircraft. This methodology describes a process in which the aircraft that is most likely to successfully complete a mission is determined computationally, and after the mission, computational methods are used to update the state of the aircraft. Once the simulation results are validated through the use of a validation tree (a specialized form of a decision tree), the new, updated model configuration is ready to be used for determining the probability of success in the next mission. Validation is an important step in performing computational simulations before the results can be fully trusted. When simulating aircraft under the presence of uncertainty, it is nearly certain that the final simulation results will not match exactly with the benchmark results used for validation. Therefore, the question for validation is not whether the simulation and experimental results are the same, but how close they have to be for the simulation to be considered good enough. The guideline for ???close enough??? can change depending on the consequences of falsely determining a simulation to be valid or the amount of risk the user is willing to accept. In this thesis, the validation process is completed through use of a validation tree that develops a final measure for the confidence of the validity of the simulation based on the accuracy of the results and the potential consequences of a false validation.The validation tree assumes that the simulation of the aircraft is done with uncertain input parameters. These uncertainties are propagated through the model using a sample based uncertainty quantification method. A methodology, called ROCUQ, is shown that is able to reduce the number of full-scale, fluid-structure interaction simulations that need to be run while still providing an adequate characterization of final results. This reduction is done through the use of a reduced-order model and a clustering technique that groups together sample sets that are likely to produce similar results. Only a predetermined number of representative samples from each of the clusters are chosen to be run in the full-scale simulation model and interpolation is used to complete the final distribution.