Data-Driven Modeling of Unsteady Aerodynamics Using Constrained Optimization and Random Input

A data-driven methodology is proposed to identify linear unsteady aerodynamic reduced-order models (ROMs) for flexible structures. Constrained gradient-based optimization extracts aerodynamic stiffness, damping, and time-domain convolution kernels directly from input-output data. Training data is generated efficiently via a single computational fluid dynamics (CFD) simulation using simultaneous band-limited white noise excitation across all structural modes. A custom loss function enforces physical kernel smoothness and temporal decay, while a spectral coherence metric assesses data linearity to establish excitation bandwidth limits. Validated against linear potential flow and coupled Euler flow simulations for a clamped plate in supersonic flow (Mach 1.2–1.94), the extracted ROMs accurately predict flutter onset and post-flutter limit-cycle oscillations. The approach demonstrates high scalability, successfully identifying a 16-mode, 50,000-parameter system where classical least-squares methods fail. Finally, the method is applied to a plate with an impinging oblique shock to compute nonlinear aeroelastic stability boundaries, which are validated via direct comparison with coupled Euler flow solutions.

This work is towards an M.Sc. degree under the supervision of Assoc. Prof. Maxim Freydin, Aerospace Engineering, Technion.