Flutter Boundary Prediction Method Based on Long Short-Term Memory Networks

To effectively predict the flutter critical velocity in the subcritical speed range, this study focuses on the analysis of the turbulent natural excitation response signals of wings in wind tunnel flutter tests. A flutter boundary prediction method is proposed using Long Short-Term Memory (LSTM) networks for flutter feature extraction from the time-frequency spectra of turbulent excitation response signals. The response data is transformed into the time-frequency domain through continuous wavelet transform, and LSTM networks are employed to establish a classification model for response data under different experimental conditions. Additionally, regression models are developed for response data at different wind speeds under the same experimental condition. By combining the two models and analyzing the flutter data, weighted calculations are performed to estimate the current flutter state, thereby predicting the critical velocity. The results demonstrate that compared to traditional flutter boundary prediction methods that utilize modal parameters, as well as flutter estimation methods employing spectral features and conventional classification models, the proposed approach enhances robustness of flutter estimation. Additionally, it improves estimation accuracy, thereby reducing the risk associated with flutter wind tunnel tests and flight tests.