Prediction of Subsonic Flutter Speeds for Composite Missile Fins Using Machine Learning

In the present paper, using machine learning techniques (ML), the prediction of flutter speeds for composite material missile fins is estimated for the subsonic flight regimes. Several types of composite materials are investigated (Kevlar, carbon, and E-glass S) for different fin geometries that are used nowadays in the aerospace industry as a building block for fly-worthy primary structural components. For flutter speeds data collection, the hybrid methodology is deployed, experimental, by performing tests in the subsonic wind tunnel, whereas the synthetic data for the ML model is generated using modified NACA flutter boundary equations. These equations were modified for orthotropic materials manufactured in the form of thin-walled structures. Based on this dataset, several algorithms were analyzed in order to create the ML flutter model, and it was found that for the problem on hand, the LightGBM regression approach renders the most accurate results (max coefficient of determination values) when compared to other investigated algorithms. Using ML Net technologies, a bespoke flutter software was developed. The results obtained were compared to the results obtained by experiments, known flutter binary models, and finally, numerical models based on structural, aerodynamic spline finite element approach. A good agreement between flutter speeds models was obtained, and it was concluded that the ML approach based on the LightGBM regression can be successfully used for problems where the subsonic flutter speeds estimate for composite missile fins is sought.