Nonlinear Flight Control Of Quadrotor UAVs:Research And Implementation

The researches in this article are mainly focusing on the accurate tracking and stabile control of quadrotor UAVs.Controller design is complicated considering quadrotor’s inherent system features such as nonlinearities,static instabilities,underactuation,and cross couplings.To study the nonlinear control techniques for quadrotors,the nonlinear kinematics relation,nonlinear dynamics and the actuating system are firstly modeled and identified.The principles and characteristics of the actuating system are firstly studied,and an Extreme Learning Machine(ELM)based real-time actuating system fault detecting method is proposed.In order to deal with the model nonlinearities and the full control problem,a nonlinear controller is deduced via the Command Filtered Backstepping(CFBS)technique,under the control of which the system is globally asymptotically stabilized.Further,a novel parameter-scheduling scheme is proposed,aided by which the control quality is enhanced during actual implementation.In order to reduce the demand for precise model as well as to reduce complexities in parameter tuning,a learning based control approach is explored,an ELM aided online trained attitude stabilizer is proposed and tested on an actual quadrotor.The main contributions include:(1)Focusing on miniature quadrotor UAVs,a system model including nonlinear kinematic relations,nonlinear dynamics and actuating system is constructed.The detailed identification techniques are introduced and discussed.The dynamic model related physical parameters are measured by experiment.And the actuating system related parameters are acquired by identification experiments.A model verification experiment is conducted,results show that the system model is of high precision.The modelling procedures including parameter measuring and identification techniques are innovative and practical,and can be generalized to model multi-rotor UAVs of various forms.(2)Specific to the demand of real-time actuator fault detection,an actuator fault detecting method based on ELM is proposed.An ELM network is used to approximate the dynamics of the actuator,which is trained by the actuator’s data including inputs,outputs and states.After training,the dynamics of the actuator are mapped to the output matrix of the proposed ELM network.By monitoring the norm of the output matrix,the dynamics are supervised,and system faults can be detected.The proposed ELM network is implemented and tested on an experimental propeller system platform,results show that the method is sensitive and effective on both major and minor fault circumstances.(3)A nonlinear controller for a quadrotor helicopter is proposed,the controlled quadrotor system is globally asymptotically stabilized with good control quality.The proposed controller is synthesized by CFBS method with a novel parameter-scheduling scheme.By scheduling controller parameters within the stabilizing region,the convergence speeds of errors in each step are adaptively adjusted based on different flight conditions.Benefits are:1)Amplitudes of control signals are reduced during fast tracking progress avoiding actuator saturation,which is hardly modeled and may cause instability.2)The controller can be more aggressively tuned to achieve better regulation accuracy.To validate the proposed method,experimental flight tests are conducted.Results comparing to PID and dynamic surface control are demonstrated,showing that the proposed controller is practical to an actual quadrotor system and can achieve good control performance.(4)Cascaded PID is the most popular technique in quadrotor control,however control performance largely relies on parameters tuning quality.In order to reduce parameters turning difficulty,an ELM based online trained attitude stabilizer is proposed.The ELM network is applied to learn the inverse model of the quadrotor’s angular subsystem with cross couplings among axes.As the network is trained online using actual flight data,no precision prior model of the controlled quadrotor is required.After training,the ELM network is used as a direct inverse controller,and the control outputs of the PID controller and the proposed ELM network are summed to control the quadrotor.To validate the practicability and effectiveness of the proposed method,the proposed controller is actual implemented and tested,results show that the proposed controller can stabilize a quadrotor UAV with reduced inner-loop tracking errors and tracking time lags.(5)Nonlinear flight control techniques are hardware dependent during implementation to actual systems.Flight control hardware platforms base on DSP and ARM architectures are respectively designed and tested,the solutions are discussed in detail.Comparisons in views of system costs,interconnection capabilities,and computational capabilities are given by actual tests.Focusing on applications with unsteady supply voltage,such as hybrid power UAVs,an Electronic Speed Controller(ESC)with supply voltage compensation is designed and implemented.By applying the proposed ESC,relative steady thrust forces can be produced by the actuating system with the presence of voltage instability.