| With the technology progress of sensors and microprocessors,micro unmanned aerial vehicle systems have made a great development and show great potential in air-filming,aerial mapping,accurate agriculture etc.MAV for delivery is one of the most urgent demand industries.Large internet e-commerce enterprises like Amazon and JD have deployed their aerial delivery strategy and have started to research and experiments.The background for this thesis is exactly the delivery drone project for JD X department.Delivery and cargo MAVs are different from ordinary consumer drones.They have higher system requirement in supervisory policy,operation mode and reliability.In control system,we have to face multi-disturbances,constrained positions and made it fully autonomous.Therefore,reliability,robustness and safety are keys to apply them in practical delivery.This thesis focuses on the study of artificial neural network(ANN)algorithms on some control systems design for MIMO nonlinear systems like the MAV system.Intelligent control algorithms based on artificial neural networks are the most important part in control system design.As an essential branch of artificial intelligence,neural network can handle uncertainties and high nonlinearities well since its abilities such as parallel processing,smooth nonlinear function approximation and self-organize study.MAV is a typical MIMO under actuated nonlinear system.After analyzing the dynamics of MAV,this thesis mainly studied the control system design for delivery MAV with considering the internal and external disturbances.Practical problems such as attitude inverse control,constrained position control,flying pendulum and chaotic secure communication are involved.Thesis main research contents are as follows.1)The development of JD delivery drone project is summarised and the reason for choosing electric multi-rotor as flying platform is explained.Dynamics analyzing and modelling are made for the JD Y6 drone.The cascade control scheme design and control performance are demonstrated.2)Inverse control of MAV attitude system is studied on the basis of neural networks.The basic idea is identifying the system's inverse model via neural networks.By improving online learning algorithm,introducing extended state observer and state-error feedback controller design,the control performance is improved.The convergence of the observer and the closed loop system have been proofed.Simulations based on data from a practical Y6 delivery drone validate the proposed control strategy.3)Control of a class of position output constraint for delivery MAV systems is studied.By integrating neural network and barrier Lyapunov function method,adaptive parameter update law and state feedback controller based on ideal weight vector norm estimation are achieved.On the other hand,two-loop cascade control architecture(position-angular speed)is adopted on order to improve the system bandwidth.Simulations and practical flight tests are conduct to verify the feasibility and effectiveness of the proposed control method.4)The interesting flying inverted pendulum is studied which is more complex than upwards transport and suspended loading for MAV system.A three-loop cascade ADRC controller is proposed based on online estimation of time-delay by neural network.A simulation platform of 3D mechanical systems is used to verify the control performance.Then,a real quadrotor is flying with an inverted pendulum to demonstrate the proposed method that not only can balance the pendulum,but also can track the reference trajectory of the flying quadrotor.5)Synchronization for delayed chaotic neural networks in high distance communication for delivery MAV systems is studied.In the presence of unknown master system's parameters and outer disturbances,an adaptive feedback controller and a controller based on sliding mode state observer are proposed to realize exponential synchronization and finite time synchronization for the slave system and the master system respectively. |
PDF Full Text Request