Effective Aerodynamics Modeling Based on Physics-Informed Neural Network and Flight Data

A high-performance aircraft aerodynamic model is a core component of the flight simulation device (FSD) and a critical function for the scenario-based airborne system validation and verification process. Thus, effective aerodynamics modeling is the key problem for the airborne system, integrated modular avionics (IMA) for example, integration and certification. In this paper, an aerodynamic modeling method was proposed, based on a neural network that makes use of the physical knowledge of computational fluid dynamics (CFD) and training data from real flight. Using another flight data as validation data, simulation analysis is conducted and the results show that (1) the accuracy performance is improved compared with the traditional methods, which means the aerodynamic characteristics model with effective prediction can be integrated into the FSD real-time environment; and (2) the method has an increased degree of generalization performance. This model was used in airborne system development, with other potential applications for pilot training, and scientific research in a D-level FSD.