Low-Reynolds-number airfoil design optimization using deep-learning-based tailored airfoil modes

• Published in: Aerospace science and technology 2022-02, Vol.121, p.107309, Article 107309
• Main Authors: Li, Jichao, Zhang, Mengqi, Tay, Chien Ming Jonathan, Liu, Ningyu, Cui, Yongdong, Chew, Siou Chye, Khoo, Boo Cheong
• Format: Article
• Language: English
• Subjects: Aerodynamic shape optimization; Deep learning; Geometric filtering; Low-Reynolds-number airfoil; Modal parameterization; UAV
• ISSN: 1270-9638; 1626-3219
• DOI: 10.1016/j.ast.2021.107309

Low-Reynolds-number high-lift airfoil design is critical to the performance of unmanned aerial vehicles (UAV). However, since laminar-to-turbulent transition dominates the aerodynamic performance of low-Reynolds-number airfoils and the transition position may exhibit an abrupt change even with a small geometric deformation, aerodynamic coefficient functions become discontinuous in this regime, which brings significant difficulties to the application of conventional aerodynamic design optimization methods. To efficiently perform low-Reynolds-number airfoil design, we present a tailored airfoil modal parameterization method, which reasonably defines the desired design space using deep-learning techniques. Coupled with surrogate-based optimization, the proposed method has shown to be effective and efficient in low-Reynolds-number high-lift airfoil design. It is found that it is necessary to consider laminar-to-turbulent transition and to perform multi-point optimization in practical low-Reynolds-number airfoil design. The maximal lift coefficient is an active constraint influencing the selection of the optimal cruise lift coefficient. The results show the complexity of low-Reynolds-number high-lift airfoil design and highlight the significance of the proposed method in the improvement of optimization efficiency.