Adaptive Gaussian Process Surrogate Modeling; Application to Flutter Boundary Uncertainty Quantification

This work presents and assesses non-intrusive surrogate modeling approaches to estimating the probability of flutter in the presence of model uncertainties for a range of flight conditions. The goal is to provide a probabilistic characterization of the flutter boundary that accounts for uncertainties in model parameters and operating conditions to better inform design and testing decisions in the certification of new aircraft. To this end, adaptively constructed Gaussian process surrogate models and high-fidelity computational aeroelasticity models are used to efficiently characterize sensitive features like the transonic dip while controlling the cost of many-query uncertainty analysis. To obtain acquisition functions for adaptive sampling, this work derives relationships between misclassification of flutter risk and existing sequential design strategie, proposes a new sampling strategy, and uses the strategies to accurately approximate flutter probability within a limited sampling budget. The efficiency of these methods are assessed using synthetic test functions, a traditional aeroelasticity model with Theodorsen’s aerodynamics model, and a computational aeroelasticity model based on the Euler equations.