The Key Technologies Research Of Moving Mass-Actuated Coaxial Dual-rotor UAV

The coaxial dual rotor aircraft has been developed rapidly in recent years due to its strong maneuverability and high flexibility.The moving mass control(MMC)mechanism changes the centroid of the UAV by controlling the movement of the internal mass sliders,thereby altering the aerodynamic torque generated by the rotors to achieve attitude control.Compared to traditional variable pitch control schemes,the MMC scheme allows the coaxial dual-rotor aircraft to avoid using easily worn variable pitch mechanisms.Additionally,the fixed-pitch rotor can provide more stable lift output,facilitating lift control.Therefore,MMC has become a current research hotspot.In this paper,the moving mass-actuated ducted coaxial rotor UAV(MAUAV)is taken as the research object.Considering its strong nonlinearity and uncertainty,structure design and optimization of MMC mechanism,the dynamics and adaptive and self-learning control are researched.Firstly,a scheme of attitude control based on three-rail MMC mechanism is proposed,which solves the problem of complex rotor components and low reliability of rotor control system.Then,the Newton-Euler modeling method is employed to construct the motion model of the three-track MAUAV,and the precise expressions of the fuselage position and the MMC mechanism’s activity are then presented.On this basis,the influence of internal disturbance caused by different mass ratios and track design layout on the maneuverability and stability of MAUAV is studied.The optimization of control mechanism is given,and it is pointed out that MMC is more effective for large size MAUAV.A comparison of the tension and attitude control efficiency of the MMC scheme and the periodic variable pitch scheme is made through aerodynamic simulation,which establishes the rotor’s aerodynamic model through the utilization of blade element momentum theory.Finally,the flow field simulation shows that the duct will provide additional lift to the rotor system.Then,the mathematical model of MAUAV is employed to derive the system state equation,and then attitude and position control are analyzed.It is showed that the characteristic of MAUAV control has nonlinearity and uncertainty.Aiming at the nonlinearity and uncertainty of MAUAV control,the adaptive position control law and attitude control law are designed by using the advantages of backstepping method in nonlinear system control and the strong robustness of adaptive sliding mode control.Simulation experiments have been conducted to verify the control effect of the designed cascade sliding mode controller in the presence of system parameter perturbation and external disturbance.Aiming at the problem that the state trajectory of sliding mode control will frequently cross the sliding mode surface and deviate from the optimal state trajectory.The design of position controller based on deep reinforcement learning is researched.The optimal state trajectory is gradually approached by training and optimizing its own network parameters.An improved method Buffer-Shared Group Optimized DDPG(BSGO-DDPG)is proposed to solve the problems of slow experience accumulation and low experience learning efficiency of Deep Deterministic Policy Gradient(DDPG)algorithm.BSGO-DDPG uses parallel training and buffer sharing to accelerate the accumulation of experience,and uses global excellent network parameter sharing to accelerate the update of each agent’s network parameters.Finally,the training environment is established,the training task is designed,and the performance of DDPG,Twin Delayed DDPG(TD3)and BSGO-DDPG algorithms is compared.It is pointed out that the position controller based on BSGO-DDPG algorithm has achieved the expected design goal in terms of performance,generalization ability and anti-interference ability.