An efficient aeroelastic analysis framework for three-dimensional wings using neural network architecture

Aeroelasticity analysis plays a critical role in ensuring flight safety and optimizing aircraft performance. However, traditional computational fluid dynamics/computational structural dynamics (CFD/CSD) coupling and wind tunnel experiments are computationally expensive and time-consuming, limiting their applicability in real-world scenarios. In this study, we focus on the aeroelastic behavior of three-dimensional (3D) wings, addressing the lack of research in this area. First, we propose an innovative data generation framework that combines modal shapes and chirp signals, enabling efficient simulation of complex dynamic deformation scenarios. Then, we introduce a method based on a convolutional neural network (CNN) and memory-in-memory (MIM) architecture for aeroelastic analysis. In this framework, CNN extracts spatial features, while MIM captures temporal dynamics. This design effectively addresses the challenges of modeling long-term dependencies and non-stationary aerodynamic behaviors faced by conventional long short-term memory models. We conduct comprehensive experiments using the Advisory Group for Aerospace Research and Development 445.6 wing, generating aerodynamic data through CFD/CSD simulations and validating the prediction performance and flutter analysis capability of the proposed model. The results demonstrate that our method achieves high prediction accuracy, reduces computational cost, and reliably identifies the critical flutter dynamic pressure. These findings validate that the proposed method is a practical solution for efficient aeroelasticity analysis of 3D wings.