Physics-Informed Transfer Learning-Based Aerodynamic Parameter Identification of Morphing Aircraft

This paper presents an aerodynamic parameter identification method for morphing aircraft through physics-informed transfer learning. The basic configuration of the morphing aircraft is first identified using physics-informed neural networks (PINN), which identifies aerodynamic parameters by integrating the flight dynamic model as a regularization term within the loss function. When the morphing aircraft transitions to the next configuration, the unidentified aerodynamic parameters change accordingly. To rapidly identify the aerodynamic parameters after deformation, a transfer learning approach is adopted. The advantages of both PINN and transfer learning are combined by this method, and a significant reduction in the required flight data for aerodynamic parameter identification can be achieved along with an improvement in model accuracy. The effectiveness of this method is demonstrated through simulation experiments. In scenarios with limited simulated flight data and noise, superior parameter identification accuracy is achieved by this method.