A physics-informed neural network-based aerodynamic parameter identification method for aircraft

The precise identification of aerodynamic parameters for aircraft is increasingly critical in aircraft design endeavors, particularly for furnishing accurate dynamic models for ground and flight simulation. To mitigate the dependence on traditional identification methods that rely heavily on aircraft models, this work introduces an aerodynamic parameters identification methodology based on Physics-informed Neural Networks (PINNs). By utilizing the six degrees of freedom motion equation as the physical constraint with the neural network, and the aerodynamic parameters to be identified as the neural network variables, the neural network model is trained to serve as a surrogate for the aircraft model. As a case study, the longitudinal motion of the aircraft is employed to identify and analyze the aerodynamic parameters. A comprehensive comparison is conducted among a method based on the genetic algorithm, a conventional neural network-based approach, and the proposed PINN-based methodology. The results obtained demonstrate that the proposed method can effectively mitigate system and data errors, exhibiting high precision in parameter identification and anti-interference capabilities with regard to noise in data. This innovative approach holds the potential to substantially decrease the reliance on flight test data for parameter identification purposes.