Scholarlink

Morphing aircrafts can optimize aerodynamic performance through dynamic shape reconfiguration during multiple flight phases, significantly enhance overall flight efficiency in complex mission scenarios. However, the deformation process induces problems such as sudden aerodynamic parameter shifts and strong nonlinearities, posing serious challenges for flight control systems. To address these issues, this study proposes a reinforcement learning-driven online self-learning intelligent controller. Requiring minimal reliance on precise mathematical models, the proposed controller online adjusts control parameters through continuous interaction with flight states, effectively accommodating substantial variations in aircraft dynamics during shape transformation and maintaining robust control performance. Firstly, the multi-body dynamic model of the morphing aircraft is established by integrating its overall characteristics and control features. The control system model is then obtained by linearizing the dynamic model under the assumption of small perturbations. Secondly, to address the cross-coupling effects between the control channels, a decoupled three-loop autopilot controller architecture with a coupling compensation mechanism is designed. Subsequently, baseline controllers for the pitch, yaw, and roll channels are constructed individually. Finally, an online self-learning control framework driven by the proximal policy optimization (PPO) algorithm is proposed. The framework formulates a Markov decision process for the flight control of morphing aircrafts, defining three key elements, including state space, action space, and reward functions. This enables online optimization of the three-channel controller parameters. The simulation results showed that the intelligent controller demonstrates rapid convergence within limited training episodes. The controller maintains precise tracking of both overload and roll angle command signals across wide-ranging flight conditions. It also exhibits strong robustness against parametric uncertainties and external disturbances.

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

Online Self-Learning Intelligent Control of Morphing Aircrafts via Reinforcement Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ziyang Zhang,
  • Xiaoqi Zhou,
  • Xugang Wang

摘要

Morphing aircrafts can optimize aerodynamic performance through dynamic shape reconfiguration during multiple flight phases, significantly enhance overall flight efficiency in complex mission scenarios. However, the deformation process induces problems such as sudden aerodynamic parameter shifts and strong nonlinearities, posing serious challenges for flight control systems. To address these issues, this study proposes a reinforcement learning-driven online self-learning intelligent controller. Requiring minimal reliance on precise mathematical models, the proposed controller online adjusts control parameters through continuous interaction with flight states, effectively accommodating substantial variations in aircraft dynamics during shape transformation and maintaining robust control performance. Firstly, the multi-body dynamic model of the morphing aircraft is established by integrating its overall characteristics and control features. The control system model is then obtained by linearizing the dynamic model under the assumption of small perturbations. Secondly, to address the cross-coupling effects between the control channels, a decoupled three-loop autopilot controller architecture with a coupling compensation mechanism is designed. Subsequently, baseline controllers for the pitch, yaw, and roll channels are constructed individually. Finally, an online self-learning control framework driven by the proximal policy optimization (PPO) algorithm is proposed. The framework formulates a Markov decision process for the flight control of morphing aircrafts, defining three key elements, including state space, action space, and reward functions. This enables online optimization of the three-channel controller parameters. The simulation results showed that the intelligent controller demonstrates rapid convergence within limited training episodes. The controller maintains precise tracking of both overload and roll angle command signals across wide-ranging flight conditions. It also exhibits strong robustness against parametric uncertainties and external disturbances.