Aerodynamic Shape Optimization with Machine Learning

Abstract

Aerodynamic shape optimization with computational fluid dynamics is an effective way to automate the aircraft design process. Especially using the adjoint method, gradient-based aerodynamic shape optimization can hold high efficiency in problems with even hundreds of geometric design variables. However, there are still some intractable challenges, such as global design and fast interactive design, calling for advanced solutions. This talk will report the recent work in addressing the unsolved challenges using machine learning techniques. The core contribution is a deep-learning-based modal parameterization that avoids the curse of dimensionality issue in aerodynamic shape optimization. Based on that progress, a breakthrough is shown in solving global aerodynamic shape design and fast interactive design problems. Besides, data-driven models based on physical knowledge are developed to handle off-design aerodynamic constraints such as low-speed performance and buffeting boundaries, which shows promise in improving aerodynamic design quality.

Speaker

Dr. Jichao Li is a Postdoctoral Research Fellow at the National University of Singapore. He received his BEng degree and PhD degree from Northwestern Polytechnical University in 2013 and 2019 respectively. His research interests include aerodynamic shape optimization and flow control.