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

In complex flight scenarios such as high-angle-of-attack (AoA) maneuvers, aircraft aerodynamic parameters exhibit significant nonlinear characteristics and strong coupling effects. To address this issue, this study proposes a nonlinear aerodynamic parameter identification method for aircraft based on Physics-Informed Neural Networks (PINN). Specifically, the equations of motion characterizing the nonlinear aircraft flight behavior are directly incorporated into the neural network’s loss function, endowing physical laws with a decisive role in model training. This approach optimizes the deep learning framework to accurately identify target aerodynamic parameters. Compared with conventional methods, the proposed technique significantly reduces dependence on large-volume training data while substantially enhancing modeling accuracy for complex nonlinear systems and the generalization capability across the full flight envelope. Flight simulation results demonstrate that under 2% root-mean-square measurement noise, the mean absolute errors and normalized root-mean-square errors for the drag coefficient are 0.0075 and 0.0435, respectively; those of the lift coefficient are 0.0092 and 0.0674, respectively; and those of the pitching moment coefficient are 0.0043 and 0.0847, respectively. These findings indicate that the PINN-based method exhibits excellent precision in identifying nonlinear aircraft parameters and maintains high identification accuracy even with flight data containing significant noise.