Scholarlink

Aiming at the difficulty of precise flight fault-tolerant control of civil aircraft under actuator fault and parameters uncertainties, an intelligent Fault-tolerant control(FTC) method based on Reinforcement Learning-Incremental Nonlinear Dynamic Inversion(RL-INDI) is proposed. The INDI control method is used to solve the model uncertainty caused by actuator fault. The reinforcement learning algorithm is used to interact with the external environment, train and adjust the parameters of the INDI controller according to the system state, and realize the precise flight fault-tolerant control. Finally, the simulation results of the Cessna550 aircraft show that the proposed RL-INDI based intelligence fault-tolerant control method has better performance. The control error is reduced by 9.2, 0.25, 1.3(rad/s) respectively, and the adjustment time is shortened by 0.6, 0.08, 1.04(s) respectively.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Reinforcement Learning-Incremental Nonlinear Dynamic Inversion Based Intelligent Fault-Tolerant Control

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Qian Zhang,
  • Shiqian Liu,
  • Weizhi Lyu,
  • Congjie Yang,
  • Jingzhou Dai Ruan,
  • Han Chen

摘要

Aiming at the difficulty of precise flight fault-tolerant control of civil aircraft under actuator fault and parameters uncertainties, an intelligent Fault-tolerant control(FTC) method based on Reinforcement Learning-Incremental Nonlinear Dynamic Inversion(RL-INDI) is proposed. The INDI control method is used to solve the model uncertainty caused by actuator fault. The reinforcement learning algorithm is used to interact with the external environment, train and adjust the parameters of the INDI controller according to the system state, and realize the precise flight fault-tolerant control. Finally, the simulation results of the Cessna550 aircraft show that the proposed RL-INDI based intelligence fault-tolerant control method has better performance. The control error is reduced by 9.2, 0.25, 1.3(rad/s) respectively, and the adjustment time is shortened by 0.6, 0.08, 1.04(s) respectively.