Massively Multipoint Aerodynamic Shape Design via Surrogate-Assisted Gradient-Based Optimization

Abstract

Multipoint aerodynamic design optimization involves no more than tens of flight conditions, which cannot thoroughly represent the actual demand. A comprehensive evaluation of the performance may consider hundreds or even thousands of flight conditions, and this leads to a massively multipoint optimization problem. Existing optimization methods are inefficient in such cases. This paper presents a surrogate-assisted gradient-based optimization architecture that efficiently solves massively multipoint design problems. To avoid the curse of dimensionality, surrogate models are constructed only in the low-dimensional space spanned by flow condition variables. With the aerodynamic functions and gradients computed by surrogate models, efficient gradient-based optimization is performed to find the optimal design. To ensure convergence, an adaptive sampling criterion is proposed to refine the surrogate models. In a transonic aircraft wing design case, the results show that the optimal design found by the proposed method with 342 missions yields a fuel burn reduction by a factor of two as compared to a regular multipoint optimal design. This work highlights the demand and provides an efficient way to conduct massively multipoint optimization in aircraft design.