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

The article provides a comprehensive review of aerodynamic shape optimization (ASO) using deep learning surrogate models, emphasizing recent advances, key technologies, and challenges. It contrasts traditional computational fluid dynamics (CFD) methods with data-driven neural networks (e.g., CNNs, GANs) and physics-driven neural networks, including Physics-Informed Neural Networks (PINNs), Deep Operator Networks (DeepONets), and Fourier Neural Operators (FNOs), highlighting their principles, applications, and comparative advantages. The review discusses optimization algorithms such as genetic algorithms and Bayesian optimization integrated with surrogate models, and presents practical cases demonstrating improved aerodynamic performance in airfoil and blade design. It also addresses computational resource demands, model interpretability, and future prospects, including emerging Kolmogorov–Arnold Networks (KAN) and other novel neural architectures, underscoring the importance of hybrid approaches and interdisciplinary collaboration for advancing efficient, physically consistent aerodynamic optimization.