基于深度学习的四旋翼无人机地面效应补偿降落控制设计

针对四旋翼无人机在降落控制过程中地面效应对控制性能有较大影响的问题,在地面效应复杂,难以建立机理模型的约束下,提出一种基于深度学习的新型非线性鲁棒控制策略.利用深度神经网络的学习能力,建立无人机降落过程中未知地面效应的补偿模型;结合super-twisting控制设计,实现对降落过程中未知地面效应的快速抑制和无人机降落的精确控制;通过Lyapunov分析法和谱归一化法,证明降落过程中闭环系统的稳定性和无人机位置误差的有限时间收敛特性.实时飞行实验结果表明,所提出的控制策略具有较好的控制效果.     

基于HOD的无模型四旋翼RBF滑模控制

为了探索解决在无模型控制算法中如何对系统的未知模型和扰动进行准确估计,提出一种基于高阶微分器(HOD)的无模型RBF神经网络滑模控制器(HODRBFSMC).引入HOD估计系统模型的各阶状态变量,并将系统模型的未知项和外界干扰统一归为总扰动,通过RBF神经网络对总扰动进行估计,并根据Lyapunov定理证明所设计控制器的闭环稳定性.为验证控制器的有效性,所设计的控制器被应用于四旋翼飞行器的轨迹控制,解决其模型参数复杂且飞行过程中易受外界干扰的问题.仿真实验表明,所提出方法能够有效估计并补偿总扰动,其轨迹跟踪能力和抗干扰性能相比PID和高阶微分反馈控制(HODFC)具有一定的优越性,能够很好地满足四旋翼飞行器控制的需求.     

Control Surface Faults Neural Adaptive Compensation Control for Tailless Flying Wing Aircraft with Uncertainties

A neural adaptive compensation tracking control scheme considering the prescribed tracking performance bound is proposed for a flying wing aircraft with control surface faults, actuator saturation and uncertainties of aerodynamic parameters. Second-order command filters are introduced to avoid the saturation of the actuators, prescribed performance bound strategy is designed to characterize the convergence rate and maximum overshoot of the tracking error, uncertainties of aerodynamic parameters are approximated by online RBF neural networks, and control allocation law is designed to reduce the coupling of the flight dynamics. The closed-loop control law is given based on adaptive backstepping compensation control scheme, and the stability of the closed-loop system is proved by Lyapunov based design. Simulation results are given to illustrate the effectiveness of the proposed neural adaptive compensation control scheme.     

无人机编队飞行神经网络自适应逆控制器设计

针对无人机编队飞行时存在的气动耦合和外部干扰等影响因素,提出基于“长-僚机”模式的神经网络自适应逆控制器设计方法.详细推导了气动耦合影响,建立了完整的编队飞行非线性数学模型,设计了非线性动态逆控制律,提出了改进的 BP 神经网络算法,自适应地逼近和在线补偿动态逆误差,改善了控制效果,并针对队形变换提出了简单有效的设计思想.仿真表明,该控制器能有效实现编队队形的保持或变换,控制系统结构具有良好的扩充性.     

Adaptive Neural Network Control of Small Unmanned Aerial Rotorcraft

This paper proposes an online learning adaptive neural network for small unmanned aerial rotorcraft to improve control performance during flight. Based on state error information, the weight matrix of the adaptive neural network can be updated on line by using lyapunov function. Therefore, no prior training data is needed for the training of the adaptive neural network. Combined with feedback control, the adaptive neural network can construct the map between the state error information and disturbances to compensate for system disturbances. The effectiveness of the proposed method is validated by a series of simulations and flight tests. Compared with feedback control method, the adaptive neural network control method can estimate and eliminate disturbances quickly to yield a good tracking performance.     

Tracking of a moving ground target by a quadrotor using a backstepping approach based on a full state cascaded dynamics

In this paper, a tracking controller is formulated for a quadrotor to track a moving ground target. The quadrotor exhibits distinct hierarchical dynamics that allows its position to be controlled by its attitude. This motivates the use of backstepping control on the underactuated quadrotor. Most backstepping architecture controls the quadrotor position and attitude independently, and couples them with inverse kinematics. Inverse kinematics computes the attitude angles required to achieve a desired acceleration. However unmodeled effects are shown to cause inexact inversion resulting in tracking error. The approach proposed in this paper uses a re-formulated full state cascaded dynamics to eliminate the need for inverse kinematics in a full state backstepping control architecture. It is shown that zero steady state error is achieved in the presence of unmodeled aerodynamics effect and wind disturbance despite no integral action. In addition, a backstepping formulation is derived using contraction theory that guarantees the boundedness of state response under bounded disturbances such as wind. This improves the system performance. Numerical simulations are performed using the proposed controller to track a target moving along predefined paths and the results are compared with a benchmark controller derived using inverse kinematics. The results show that the proposed controller is able to achieve better tracking performance under unmodeled aerodynamic effects and wind disturbance as compared with the benchmark controller.     

基于神经网络观测器的飞控传感器在线故障诊断

为提高无人机飞行安全可靠性,针对飞行控制系统中常出现的传感器故障以及非线性气动力模型参数难以确定的问题,提出了基于BP神经网络观测器估计的故障诊断方法;引用LM改进算法对网络参数进行调整,构造了神经网络观测器模型逼近非线性系统,并运用于飞行控制系统进行在线数字仿真,对垂直陀螺输出卡死故障、恒偏差故障和恒增益故障分别进行仿真分析;仿真结果表明,所设计神经网络观测器可以有效估计系统输出,在线诊断传感器故障。     

无人机吊挂空运系统的自适应控制设计

针对四旋翼无人机吊挂空运系统存在的模型不确定性及欠驱动性问题,本文提出了一种基于能量耦合的自适应控制设计.首先,基于能量整形控制方法构造了一种新型的能量存储函数以处理状态耦合.然后利用神经网络对系统未建模动态特性进行在线估计,同时设计参数自适应律在线估计模型中的未知参数,并采用基于符号函数的鲁棒控制算法补偿神经网络的估计误差.本文运用李雅普诺夫方法和拉塞尔不变性原理对闭环系统的稳定性进行了证明,并且证明了负载摆动和无人机位置误差的渐近收敛性.最后,在室内实验平台上进行了飞行实验.实验结果表明,本文提出的非线性控制方法能够在有效抑制吊挂负载摆动的同时,实现无人机位置的精确控制.     

推力矢量可倾转四旋翼自抗扰飞行控制方法

针对常规四旋翼难以实现位置和姿态独立控制问题, 研究了一种具有全向推力矢量的可倾转四旋翼飞行 器系统. 为克服系统的大范围不确定性、强耦合性及外部风扰影响, 设计了基于自抗扰控制(ADRC)技术的飞行控 制器. 通过建立风扰下的系统动力学模型, 分析阵风对旋翼气动力的影响. 接着将系统解耦为六通道单回路结构并 分别设计自抗扰控制器, 引入扩张状态观测器估计系统的内外扰动, 利用非线性状态误差反馈律输出扰动补偿控 制. 在此基础上, 通过变量代换线性化控制分配矩阵, 将控制器输出直接映射到旋翼转速和倾转角. 仿真结果表明, 所设计的自抗扰飞行控制器具有良好的位置和姿态独立控制能力, 能够有效地估计和补偿紊流风扰动, 同时对系统 的部分动力失效故障有较强的鲁棒性.     

基于多传感器的四旋翼飞行器硬件系统设计

四旋翼飞行器由于其简单的机体结构与较为复杂的姿态控制,近年来在军用和民用领域广泛应用。旨在通过四旋翼飞行器飞控平台的搭建,实现对飞行器姿态的稳定控制。首先论述了四旋翼飞行器的飞行原理与机械结构,给出了硬件系统总体结构。在对各功能模块整合的基础上,实现基于多传感器的控制系统硬件电路设计。仿真与实验证明：多传感器使用过程中,通过卡尔曼滤波进行姿态数据的融合,有效地解决了加速度计、陀螺仪易受外界干扰问题,所设计硬件系统在飞行实验中性能稳定,为四旋翼的稳定控制提供了参考。     

基于神经网络的故障飞机仿真

传统气动系数模型中，拟合法精度较差，插值法计算速度慢，且占内存多。利用神经网络一致逼近任意非线性连续函数的特性，训练具有一个三输入六输出的神经网络模型，建立故障飞机仿真系统。仿真结果和故障飞机自修复应用表明，文中所采用的神经网络建模方法是可行的。在自修复飞行控制系统研究中，为故障飞机建模所需大量故障状态气动系数数据处理提供一种新思路。     

自适应神经网络四旋翼无人机有限时间轨迹跟踪控制

针对带有模型不确定性和未知外部干扰的四旋翼无人机轨迹跟踪控制问题,提出一种基于径向基(radial basis function, RBF)神经网络的自适应全局快速终端滑模控制方法,确保系统对期望轨迹的有限时间跟踪。该方法考虑到全局快速终端滑模控制在实际应用中的适应性和抖振问题,利用RBF神经网络替代等效控制量,以神经网络的在线学习能力补偿系统内部的不确定性和未知的外部干扰,有效地降低了系统的抖振;根据Lyapunov方法导出的自适应律在线调整神经网络权值,以保证闭环系统的稳定性。通过一系列仿真算例和飞行实验验证了该方法的有效性与可行性,结果表明:该控制方法相对于滑模控制的抖振更小,具有更好的收敛性和抗干扰能力,同时对模型的参数摄动具有更强的鲁棒性。     

四旋翼无人机飞控系统的研究与实现

本文通过数学建模进行动力学系统分析,研究实现了基于硬件和软件的四旋翼无人机飞控系统。首先、构建了四旋翼无人机动力学模型并进行理论分析;其次、设计了无人机机架,对各组成模块进行测试、分析和试验;再次、通过集成软硬件实现了无人机飞控系统并进行飞行测试;最后、实验结果表明,实现的无人机飞控系统取得了较好的飞控效果,具有灵敏性强、稳定性高,总体性能优良等优点。     

Flight envelope protection of aircraft using adaptive neural network and online linearisation

Flight envelope protection algorithm is proposed to improve the safety of an aircraft. Flight envelope protection systems find the control inputs to prevent an aircraft from exceeding structure/aerodynamic limits and maximum control surface deflections. The future values of state variables are predicted using the current states and control inputs based on linearised aircraft model. To apply the envelope protection algorithm for the wide envelope of the aircraft, online linearisation is adopted. Finally, the flight envelope protection system is designed using adaptive neural network and least-squares method. Numerical simulations are conducted to verify the performance of the proposed scheme.     

基于强化学习的四旋翼无人机控制律设计

目前四旋翼无人机大部分都采用经典控制方法进行控制律的设计,然而控制参数的选择和对被控对象数学模型的依赖一直是经典控制方法设计中需要克服的问题;针对此问题,采用了一种基于深度强化学习算法Deep Q Network的无人机控制律设计方法,以四旋翼姿态角和姿态角速率作为三层神经网络的输入数据,最终输出动作值函数,再根据贪婪策略进行动作的选取,通过与环境的不断交互,智能体根据奖惩信息来更新神经网络的权值,使得智能体朝着获得累积回报最大值的方向选取动作;仿真结果表明在经过强化学习训练之后,四旋翼姿态角能够快速准确地跟踪上参考指令的变化,证明了基于强化学习的四旋翼无人机控制律的可行性,从而避免了传统控制方法对控制参数的选择与控制模型的依赖。     

基于AFSMC算法的四旋翼飞行器控制策略研究

针对四旋翼飞行器在飞行过程中,控制系统存在非线性、强耦合、不确定性和鲁棒性差的问题,建立了关于四旋翼飞行器的动力学数学模型,将自适应控制、模糊控制和滑模控制相结合,提出基于自适应模糊滑模控制(AFSMC)的快速平稳控制策略。采用模糊系统推理方法实现理想控制律的逼近。在满足李雅普诺夫稳定性条件的前提下进行控制器的设计和稳定性分析,并结合四旋翼的数学模型和给定参数进行了MATLAB仿真。仿真结果表明,AFSMC控制器相比常规PID控制器具有良好的动态性能和抗干扰能力。     

基于加速度补偿的无人机吊挂飞行抗摆控制

为了抑制四旋翼无人机(UAV)吊挂飞行中的载荷摆动,研究了一种新的基于加速度补偿的抗摆控制方法.首先,基于拉格朗日法建立了四旋翼UAV吊挂系统的非线性动态特性方程,并构建了能量函数来设计飞行控制系统,使四旋翼UAV跟踪参考轨迹;然后,利用吊挂载荷运动轨迹广义误差设计抗摆控制器,对四旋翼UAV进行加速度补偿以修正UAV的...     

Nonlinear robust control of a quadrotor helicopter with finite time convergence

In this paper, the control problem for a quadrotor helicopter which is subjected to modeling uncertainties and unknown external disturbance is investigated. A new nonlinear robust control strategy is proposed. First, a nonlinear complementary filter is developed to fuse the raw data from the onboard barometer and the accelerometer to decrease the negative effects from the noise associated with the low-cost onboard sensors Then the adaptive super-twisting methodology is combined with a backstepping method to formulate the nonlinear robust controller for the quadrotor''s attitude angles and the altitude position. Lyapunov based stability analysis shows that finite time convergence is ensured for the closed-loop operation of the quadrotor''s roll angle, pitch angle, row angle and the altitude position. Real-time flight experimental results, which are performed on a quadrotor flight testbed, are included to demonstrate the good control performance of the proposed control methodology.     

时滞不确定飞行系统神经网络非脆弱H∞控制

针对具有参数摄动和状态时延的时滞不确定飞行系统,提出了一种神经网络非脆弱H_∞控制方案。该方案将鲁棒H_∞控制和神经网络控制结合起来,利用径向基神经网络的非线性逼近能力,对飞行系统的非线性不确定项进行逼近。由线性矩阵不等式（LMI）设计系统标称部分的鲁棒控制器,然后利用神经网络的输出来消除系统控制输入中的不确定部分。Lyapunov稳定性分析中,综合考虑了系统参数摄动、时延和神经网络逼近误差的影响,并证明了在所设计的飞行控制器作用下,闭环系统的稳定性。仿真实例验证了提出的飞行控制方案的可行性和有效性。     

Direct adaptive neural flight control system for an unstable unmanned aircraft

A direct adaptive controller design using neural network is proposed for an unstable unmanned research aircraft similar in configuration to combat aircraft. The control law to track the pitch rate command is developed based on system theory. Neural network with linear filters and back propagation through time learning algorithm is used to approximate the control law. The bounded signal requirement to develop the neural controller is circumvented using an off-line finite time training scheme, which provides the necessary stability and tracking performances. On-line learning scheme is implemented to compensate for uncertainties due to variation in aerodynamic coefficients, control surface failures and also variations in center of gravity position. The performance of the proposed control scheme is validated at different flight conditions. The disturbance rejection capability of the neural controller is analyzed in the presence of the realistic gust and sensor noises. Hardware-in-loop simulation is also carried out to study the behavior of control surface deflections in real-time.