[home] [Personal Program] [Help]
tag
18:00
1 mins
Deep Reinforcement Learning for Aeroservoelastic Control of Unmanned Multi-Body Aircraft with Geometric Nonlinearities
Zhuolin Ying, Ying Bi, Zijian Zhu, Chen Zhu, Xiaoping Ma
Session: Poster session & drinks
Session starts: Tuesday 18 June, 18:00
Presentation starts: 18:00
Room: Room 1.1


Zhuolin Ying ()
Ying Bi ()
Zijian Zhu ()
Chen Zhu ()
Xiaoping Ma ()


Abstract:
The research delves into the unique challenges posed by high-altitude long-endurance (HALE) aircraft, particularly focusing on their extreme flexibility. The study explores the substantial deformation experienced by the lightweight and highly flexible wings of HALE aircraft during flight. This deformation not only alters aerodynamic calculation but also has a pronounced impact on the stiffness characteristics of the wings. To address issues of poor wind resistance during takeoff and landing, the paper introduces an innovative solution – an unmanned multi-body aircraft (MBA) design featuring wingtip docking. However, the introduction of multiple wingtip dockings brings about inevitable geometric nonlinearity issues, significantly influencing the aeroservoelastic design of the aircraft. Consequently, there is an urgent need for aeroservoelastic analysis for unmanned multi-body aircraft that takes geometric nonlinearity into account. The research methodology involves the establishment of a nonlinear finite element model tailored to flexible unmanned multi-body aircraft. The minimal state method is employed for rational aerodynamic approximation, allowing a comprehensive exploration of the coupling issues arising from geometric nonlinearity and aeroelasticity in the time domain. Additionally, an adaptive control law, incorporating principles of machine learning, is designed to actively suppress flutter. The findings of the research reveal that flexible unmanned multi-body aircraft considering geometric nonlinearity may exhibit complex behaviors such as limit cycle oscillations and chaotic phenomena under specific initial conditions. Importantly, the proposed adaptive learning control law emerges as a highly effective measure in reducing the critical flutter speed, offering robust support to enhance the overall flight stability of the aircraft. This comprehensive exploration contributes valuable insights into the intricate dynamics of flexible unmanned multi-body aircraft.