Session: 02-03-02: Aeroacoustics, Dynamic Loads, Wave Propagations, Response, Vibration, Control, and Alleviation of Aerospace Structures and Vehicles

Paper Number: 183095

183095 - Data-Driven Predictive Control of Flexible Wing via Parametric Dynamic Mode Decomposition 

This paper presents a data-driven predictive control for the nonlinear aeroelasticity of highly flexible wing. The data-driven model is derived via parametric dynamic mode decomposition as a linear parameter-varying reduced-order model (LPV-ROM) by directly using the data snapshots obtained at varying flight conditions. The LPV-ROM is represented as a state-space model with polynomial dependency on flight conditions. In the numerical studies, a highly flexible cantilever wing and a slender vehicle are first studied with fixed angles of attack as the scheduling parameter. The data-driven p-DMD modeling is then compared against the traditional linearization-based parametric modeling to verify the new modeling method's accuracy. In addition, the proposed p-DMD method is applied to the highly flexible aircraft in a perturbed longitudinal flight with varying angles of attack as the scheduling parameter. The nonlinear aeroelastic and flight dynamic data are compared with the simulation results of the data-driven p-DMD model.

After that, the data-driven LPV-ROM is used to design predictive control to suppress the vibrations excited by external gusts. The flaps will be used to suppress the vibrations (strains) of the flexible wing.  Simulation results will be provided to demonstrate the control performance of such strategy and comapred with other control methods. The performance metrics include magnitude and energy of strains along the wing span, and the control authority of the flaps. The proposed control will be applied to two flight conditions: fixed scheduling parameters and varying scheduling parameters.  

The novelty of this work lies in the following aspects: (1) The parametric dynamic mode decomposition renders a 'global' state-space model for nonlinear aeroelasticity, different from traditional linearization-based method at (quasi-) steady-state flight; (2) The data-driven predictive control avoids the physics-based linearized model to design the control law for vibration suppression, but relies on the data-driven model to design vibration control law under a dynamic flight conditions.  

Presenting Author: Larry Catalasan Utah State University

Presenting Author Biography: Larry Catalasan is a doctoral candidate at the department of mechanical and aerospace engineering of Utan State University. His research interest includes machine learning in modeling and control of nonlinear systems