Flutter boundary prediction method based on HHT and machine learning

A flutter boundary prediction method based on HHT and machine learning is proposed to predict the flutter velocity before the wind speed reaches the subcritical state. Natural excitation technique is used to extract impulse response signals. EMD ( empirical Mode decomposition method) is used to decompose the signal. Hilbert spectrum was obtained and analyzed by HHT to decompose the signal. The analysis methods included HHT spectrum and marginal spectrum analysis, so as to extract the characteristic quantity and establish the classification model according to different flight states. Then, regression models were established under different flutter modes for flutter degree analysis. During the prediction, according to the classification performance of the data to be measured, the flutter degree analysis result is weighted to obtain the flutter degree corresponding to the current wind speed, and then the flutter wind speed is calculated. In the selection of machine learning algorithm, naive Bayes algorithm, K-nearest neighbor algorithm and other machine learning algorithms are used to construct the classification model, linear regression,, Gaussian process regression and so on are used to construct the regression model. The results show that the K-nearest neighbor algorithm performs best in the classification algorithm, while the Gaussian process regression algorithm performs best in the regression algorithm. Through the cross-validation of the test data, the proposed method can accurately predict the critical flutter velocity when it is far away from the flutter boundary through flutter mode recognition and flutter degree analysis.