Multimission Fuel-Burn Minimization in Aircraft Design: A Surrogate-Modeling Approach.

摘要

Aerodynamic shape and aerostructural design optimizations that maximize the performance at a single flight condition result in designs with unacceptable off-design performance. While considering multiple flight conditions in the optimization improves the robustness of the designs, there is a need to develop a rational strategy for choosing the flight conditions and their relative emphases such that multipoint optimizations reflect the true objective function. In addition, there is a need to consider uncertain missions and flight conditions. In this thesis, the strategies to formulate the multipoint objective functions for aerodynamic shape and aerostructural optimization are presented. To determine the flight conditions and their corresponding weights, a novel surrogate-based mission analysis is developed to efficiently analyze hundreds of actual mission data to emulate their flight condition distribution. Using accurate and reliable surrogate models to approximate the aerodynamic coefficients used in the analysis makes this procedure computationally tractable. A mixture of experts (ME) approach is developed to overcome the limitations of conventional surrogate models in modeling the complex transonic drag profile. The ME approach combines multiple surrogate models probabilistically based on the divide-andconquer strategy. Using this model in the mission analysis significantly improves the range estimation accuracy, as compared to other conventional surrogate models. As expected, the multipoint aerodynamic shape and aerostructural optimizations demonstrate a consistent drag reduction, instead of the localized improvement by the single-point optimizations. The improved robustness in the multipoint optimized designs was also observed in terms of the improved range performance and more consistent fuel-burn reduction across the different missions. The results presented in this thesis show that the surrogate-model-assisted multipoint optimization produces a robust design that is optimized for a set of flight conditions matching real-world operations, which is ensured by the use of historical flight data.