TRANSONIC FLUTTER BOUNDARY PREDICTION OF WINGS WITH STRUCTURAL UNCERTAINTIES BASED ON POINT CLOUD ENCODER

To address the high computational cost of transonic unsteady aerodynamic calculations and the limitations of existing reduced-order models in predicting unsteady aerodynamics under varying structural properties, this paper proposes a variable structural property unsteady aerodynamic reduced-order model and flutter prediction method based on a point cloud encoder and multi-fidelity data fusion. Taking the AGARD 445.6 wing as the research object, low-fidelity unsteady aerodynamic forces are rapidly calculated based on the Doublet Lattice Method (DLM), which are then combined with high-fidelity unsteady aerodynamic forces from CFD with prescribed modal motions as label data to construct Long Short-Term Memory (LSTM) models for flutter prediction under both deterministic and uncertain structural parameter scenarios. First, a reduced-order model (ROM1) for identical structural properties with variable Mach numbers is established, achieving accurate and rapid prediction of unsteady aerodynamics and flutter boundaries across different Mach numbers from Ma = 0.78 to 0.95. Subsequently, to address structural uncertainty requirements, a point cloud encoder method was developed to extract the spatial geometric features of variable modal shapes when the wing’s modal shapes change. A reduced-order model (ROM2) was established for wings with structural uncertainty under variable Mach number conditions, and rapid and accurate predictions of flutter boundaries were carried out. The results show that the flutter speed prediction error of ROM1 on the test set for different Mach numbers is less than 8%, and the flutter speed prediction error of ROM2 on the test set for different structures and Mach numbers is less than 11%. Under structural parameter uncertainty, the prediction time for various Mach number conditions is only 0.2% of that required by the CFD method. This effectively enhances the efficiency of transonic nonlinear flutter prediction and engineering aeroelastic design when structural parameters change during the preliminary design stage.