中国科学院自动化研究所机构知识库

轻质细长结构的机械臂因其在长臂展、高机动、低能耗等方面的优势，在航天、建筑、核工业等领域的应用越来越广泛，但其运动过程中的末端振动问题对作业稳定性带来较大影响，因此，轻质细长结构的机械臂的振动控制已成为国内外研究的热点之一。研究人员将这种轻质细长结构的机械臂统称为柔性臂。目前针对柔性臂的研究工作中，研究对象主要是仅具有旋转关节的单连杆或者多连杆柔性臂，但混合结构的柔性臂更能适应复杂紧凑空间的大范围作业，其运动稳定性问题更值得进行深入研究。相比于仅包含旋转关节的单连杆或者多连杆柔性臂，同时具有旋转关节和伸缩关节的混合结构柔性臂的动力学耦合性更强、参数变量更多、振动模态更复杂、控制难度更大。本文将具有一个旋转关节和一个伸缩关节并在竖直平面内运动的混合结构柔性臂作为研究对象，并围绕系统建模、轨迹规划、状态估计、振动控制等问题开展研究。主要内容包括：
1. 结合实际应用需求，分析混合结构柔性臂的研究意义，并归纳总结了混合结构柔性臂抑振控制过程中的关键问题，在此基础上有针对性地对柔性臂的关键技术的研究现状进行了综述，提出了本文的整体研究思路。
2. 针对混合结构柔性臂刚性变量和柔性变量相互耦合的问题，提出一种基于分解思想的混合结构柔性臂建模方法，能够有效性的避免刚柔变量耦合和模型求解复杂的问题。首先设计并搭建了具有一个旋转关节和一个伸缩关节的混合结构柔性臂实验平台，确定实验平台中各个物理参数。在建模过程中，将伸缩关节分解为基础段、重叠段和伸出段，并将系统分解为刚性部分和柔性部分，并对刚柔两部分的分别进行建模分析。对于刚性部分，采用传统的刚性力学分析方法得到刚性系统模型。对于柔性部分，采用假设模态法得到柔性部分弹性偏差，并采用拉格朗日方程建立柔性系统模型。通过综合刚柔两部分模型，建立混合结构柔性臂完整的动力学模型，并对其振动特性进行了分析和讨论。
3. 针对混合结构柔性臂关节轨迹影响末端振动的问题，提出一种基于粒子群优化算法和非均匀样条插值的抑振轨迹规划方法，能够有效地减小末端振动的产生。为了提高最优抑振轨迹的搜索效率，采用分段函数选取非均匀插值点时刻，并基于五次多项式和随机正态分布函数初始化插值点。采用粒子群优化算法优化每个插值点增量，基于优化后的插值点，采用三次样条函数重新构造新的关节轨迹。在寻优过程中，将每个采样时刻的末端振幅之和作为优化目标函数，并定义权重因子协调抑制运动过程中的振动和抑制运动停止之后的剩余振动的侧重点，优化后最终得到每个关节的最优抑振轨迹。
4. 针对混合结构柔性臂末端状态无法准确获取的问题，提出一种基于滑动离散傅里叶变换和模糊逻辑自适应卡尔曼滤波的末端状态估计方法，能有效地估计末端状态并用于反馈控制。为了提高混合结构柔性臂的末端估计精度，将混合结构柔性臂系统模型分解为挠度模型和弹性振动模型，并搭建由加速度计、姿态传感器和编码器组成的状态测量系统，采用离散傅里叶变换对测量信号进行振动分析，用于替代弹性振动模型得到末端振动状态。将挠度模型和改进的模糊逻辑自适应卡尔曼滤波结合，对振动平衡点位置状态进行估计。通过末端振动状态和末端振动平衡点位置状态，可得到精确的末端状态。
5. 针对混合结构柔性臂振动控制难度大的问题，提出一种基于强化学习和滑模控制的振动控制方法，能够有效地使得末端位置精确跟踪输入，并且减小振幅。为了更好的实现末端轨迹的精确跟踪和末端的振动抑制，将控制器分解为基于名义模型的滑模控制器和基于Actor-Critic结构的强化学习控制器，分别输出用于轨迹跟踪的驱动力矩和用于振动抑制的补偿力矩。在基于名义模型的滑模控制器中，同时采用系统的名义模型和测量反馈信号，并且增加了积分滑模部分，提高了系统的鲁棒性。在基于Actor-Critic结构的强化学习控制器中，采用基于神经网络的近似方法，并将经验优先回放方法引入其中，提高了神经网络训练效率。
最后，总结本文所取得的研究成果，并对下一步研究提出展望。

Because of its advantages in long armspread, high mobility and low energy consumption, light and slender structure manipulators are widely used in aerospace, construction, nuclear industry and other fields. However, the tip vibration problem has great influence on the stability of operation. Therefore, the vibration control of light and slender structure manipulator has become a hot research topic at home and abroad. This light and slender structure manipulator is called flexible manipulator by researchers. At present, the research object of flexible manipulator is mainly single link or multi-link flexible manipulator with only rotating joints. But the hybrid-structured flexible manipulator can adapt to a wide range of tasks in complex and compact space, and its motion stability is worth further study. Compared with the single link or multi-link flexible manipulator containing only rotating joints, the hybrid-structured flexible manipulator with both rotating joints and telescopic joints has stronger dynamic coupling, more parameter variables, more complex vibration modes and more difficult control. In this paper, a hybrid-structured flexible manipulator with a rotating joint and a telescopic joint is taken as the research object, and the problems of system modeling, trajectory planning, state estimation and vibration control are studied. The main contents are as follows:
1. Combined with practical application requirements, the research significance of hybrid-structured flexible manipulator is analyzed, and the key problems in vibration suppression control process are summarized. On this basis, the research status of key technologies of flexible manipulator is reviewed, and the research idea of this paper is proposed.
2. In view of the problem of coupling between rigid variable and flexible variable of hybrid-structured flexible manipulator, a modeling method based on decomposition idea is proposed, which can effectively avoid the problem of rigid flexible variable coupling and the difficulty of model solving. Firstly, a flexible manipulator experimental platform with a rotating joint and a telescopic joint is designed and built, and the physical parameters of the experimental platform are determined. In the modeling process, the telescopic joint is decomposed into foundation segment, overlap segment and extension segment, and the system is decomposed into rigid parts and flexible parts. The two parts of rigid and flexible parts are modeled and analyzed respectively. For rigid part, rigid system model is obtained by traditional rigid mechanical analysis method. For the flexible part, the elastic deviation of the flexible part is obtained by the assumed modal method, and the flexible system model is established by the Lagrange equation. By integrating the rigid flexible two part model, a complete dynamic model of the hybrid-structured flexible manipulator is established, and its vibration characteristics are analyzed and discussed.
3. In view of the problem of tip vibration affected by the joint trajectory of hybrid-structured flexible manipulator, a vibration suppression trajectory planning method based on particle swarm optimization algorithm and non-uniform spline interpolation is proposed, which can effectively reduce the tip vibration. In order to improve the search efficiency of optimal vibration suppression trajectories, piecewise functions are used to select non-uniform interpolation points, and interpolation points are initialized based on five polynomial and random normal distribution function. Particle swarm optimization algorithm is used to optimize the increment of each interpolation point. Based on the optimized interpolation points, the new joint trajectories are reconstructed by the three spline function. In the optimization process, the sum of the tip amplitude of each sampling time is taken as the optimization objective function, and the weight factor is defined to coordinate the suppression of the vibration in the motion process and suppress the residual vibration after stopping the motion. Finally, optimal vibration suppression trajectories of each joint are obtained.
4. In view of the problem that the tip state of hybrid-structured flexible manipulator cannot be accurately obtained, a tip state estimation method based on sliding discrete Fourier transform and fuzzy logic adaptive Kalman filter is proposed, which can effectively estimate the tip state and be used in the feedback control process. In order to improve the accuracy of the tip estimation of the hybrid-structured flexible manipulator, the system model is decomposed into a deflection model and an elastic vibration model, and a state measurement system consisting of accelerometer, attitude sensor and encoder is built. The discrete Fourier transform is used to analyze the vibration of the measured signal, and is used to replace the elastic vibration model to obtain the tip vibration state. The deflection model and the improved fuzzy logic adaptive Kalman filter are combined to estimate the vibration equilibrium position. The accurate tip state can be obtained through the state of the tip vibration and the equilibrium position.
5. In view of the difficulty of vibration control of hybrid-structured flexible manipulator, a vibration control method based on reinforcement learning and sliding mode control is proposed, which can effectively track the input position of the tip position and reduce amplitude. In order to track the tip trajectory accurately and suppress the tip vibration, the controller is decomposed into a sliding mode controller based on nominal model and a reinforcement learning controller based on Actor-Critic structure. The driving torque for trajectory tracking and the compensation torque for vibration suppression are respectively obtained. In the sliding mode controller based on nominal model, the nominal model and the measurement feedback signal are used, and the integral sliding mode part is added to improve the robustness of the system. In the reinforcement learning controller based on Actor-Critic structure, the approximate method based on neural network is adopted, and the prioritized experience replay method is introduced to improve the efficiency of neural network training.
Finally, the research results obtained in this paper are summarized, and the future research is prospected.