航空学报 > 2025, Vol. 46 Issue (19): 531850-531850   doi: 10.7527/S1000-6893.2025.31850

基于物理信息神经网络的飞机姿态预测模型

张玉刚1,2(), 杨哲1,3, 何森朋1,3, 杨文青1,2   

  1. 1.西北工业大学 航空学院,西安 710072
    2.飞行器基础布局全国重点实验室,西安 710072
    3.中国飞行试验研究院,西安 710089
  • 收稿日期:2025-01-27 修回日期:2025-04-21 接受日期:2025-06-16 出版日期:2025-06-30 发布日期:2025-06-27
  • 通讯作者: 张玉刚 E-mail:zhangyugang@nwpu.edu.cn
  • 基金资助:
    民机专项(MJZ5-1N22)

Aircraft attitude prediction model based on physical information neural networks

Yugang ZHANG1,2(), Zhe YANG1,3, Senpeng HE1,3, Wenqing YANG1,2   

  1. 1.School of Aeronautics,Northwestern Polytechnical University,Xi’an 710072,China
    2.National Key Laboratory of Aircraft Configuration Design,Xi’an 710072,China
    3.Chinese Flight Test Establishment,Xi’an 710089,China
  • Received:2025-01-27 Revised:2025-04-21 Accepted:2025-06-16 Online:2025-06-30 Published:2025-06-27
  • Contact: Yugang ZHANG E-mail:zhangyugang@nwpu.edu.cn
  • Supported by:
    Special Research on Civil Aircraft(MJZ5-1N22)

摘要:

为解决飞机试飞过程中由于超出飞行包线而导致的试飞安全事故问题,降低事故风险,根据飞机当前状态信息和舵面偏转的角度数据,预测飞机后续姿态演化过程,为辅助飞行员决策提供依据。结合飞行动力学方程与物理信息神经网络(PINNs)方法,构建了飞机姿态实时预测(FD-PINN)模型,解决了基于神经网络(NN)的飞机姿态预测模型预测精度不足、泛化能力差的问题。考虑大气环境参数的随机性和飞机操纵输入的不确定性条件下,通过FlightGear获取飞行仿真数据对模型进行了验证,计算结果表明FD-PINN模型比NN模型泛化能力更强,预测精度更高,其中迎角预测结果的均方误差降低了68.5%。

关键词: 姿态预测, 物理信息神经网络, 飞行动力学, 试飞安全, 飞行试验

Abstract:

In order to avoid the flight test safety accidents caused by exceeding the flight envelope during flight test, and reduce the accident risk, the subsequent attitude evolution process of the aircraft was predicted according to the current state information of the aircraft and the angle data of rudder deflection, so as to provide a basis for pilot decision-making support. Combining flight dynamics equation and Physical Information Neural Networks (PINNs), a real-time aircraft attitude prediction model, Flight Dynamics-Physics Informed Neural Network (FD-PINN), was constructed to solve the problem of insufficient prediction accuracy and poor generalization ability of the Neural Network-based (NN)aircraft attitude prediction model. Considering the randomness of atmospheric environmental parameters and the uncertainty of aircraft control inputs, flight simulation data were obtained by FlightGear to verify the model. The calculation results show that FD-PINN model has stronger generalization ability and higher prediction accuracy than NN model, and the mean square error of angle of attack prediction results is reduced by 68.5%.

Key words: attitude prediction, physical information neural networks, flight dynamics, flight test safety, flight test

中图分类号: