Review of deep learning-based aerodynamic shape surrogate models and optimization for airfoils and blade profiles

Abstract

In recent years, deep learning technology has developed rapidly and has shown great potential in the optimization of complex systems. In aerodynamic shape optimization, traditional computational fluid dynamics and experimental methods are limited due to issues of efficiency and cost. In contrast, deep learning surrogate models have gradually become a new alternative to traditional methods due to their advantages in nonlinear modeling, efficient computation, and flexible design. These models offer novel approaches to optimization through methods such as data regression, automatic differentiation, and operator learning. This paper presents a comprehensive review of the latest research progress in the field of aerodynamic shape optimization based on deep learning surrogate models, focusing on key technologies, application cases, and future development trends. The article first elaborates on the importance and development context of airfoil and blade profile optimization, introducing the research background and motivation. Then, it discusses the key technologies and challenges faced in aerodynamic shape optimization. Subsequently, it introduces in detail the application of deep learning as a surrogate model, including data- and physics-drisven neural networks, such as Physics-Informed Neural Networks and Deep Operator Networks, and presents practical application cases of these networks in aerodynamic shape optimization. Finally, the article looks into the future of aerodynamic shape optimization, pointing out the advantages of Kolmogorov–Arnold Networks in improving model accuracy and interpretability, as well as the potential of new types of neural networks in aerodynamic optimization, and summarizes their development.