基于运动学模型的机械臂迭代学习控制

发布时间：2018-06-24 11:50

本文选题：机械臂系统 + 运动学控制器　；参考：《浙江大学》2017年硕士论文

【摘要】：在轻工业、手工业等行业的生产过程中,存在着大量高强度的重复性工作,这类工作适合使用机械臂系统代替人类完成。在汽车、电子等行业中,机械臂已经被大量投入使用,相关的机械臂系统控制技术也已比较成熟;但是,现有的成熟机械臂系统大多使用国外知名厂商生产的精密机械臂,系统的购买和维护成本很高,我国中小型企业难以负担设备引入的高额成本。因此,选取合适的低成本机械臂系统、提出适合于实际工况的机械臂控制方案,对于提升中小型轻、手工业企业的自动化程度有很大的实际意义。本文选取一类配备运动学控制器的机械臂系统作为研究对象,设计了三种不同形式的迭代学习控制器用以解决不同情况下的控制问题,为这一类机械臂系统的控制器选取问题提供了一套比较完整的解决方案。本文的主要研究内容如下:(1)针对机械臂对象模型未知的情况,选取了一种无模型的PD型闭环迭代学习控制算法。首先,由于迭代学习算法本身非常适合解决非线性系统轨迹跟踪问题,因此选取迭代学习控制作为机械臂系统的基本控制方案;之后,本文对比了闭环迭代学习算法与开环算法的特点,选取了对于环境干扰抑制效果较好的闭环形式迭代学习算法;最后,考虑到PID形式的控制器结构简单、鲁棒性强的特点以及迭代学习过程本身在迭代轴上的积分效应,选取了PD型闭环迭代学习律作为机械臂系统的控制算法。仿真实验证明了该算法对于解决机械臂轨迹跟踪问题的有效性。(2)针对机械臂对象参考模型已知的情况,提出了一种基于固定运动学模型的迭代学习控制算法。通过使用机械臂对象的参考模型作为先验知识,利用运动学逆解得到的参考关节角指导迭代学习过程,该算法能够有效加快系统跟踪误差的收敛速度。仿真结果显示,通过使用基于固定运动学模型的控制算法,系统误差收敛速度明显提升。(3)针对机械臂参考模型与实际对象偏差较大的情况,提出了一种基于自适应模型的迭代学习控制算法。该算法使用卡尔曼滤波方法对机械臂运动学模型参数进行在线估计,每次迭代后更新对象的参考模型用于下一步迭代学习控制。仿真实验结果表明,通过使用基于自适应模型的算法,机械臂系统在对象参考模型失配较为严重的情况下依然能够顺利完成轨迹跟踪的任务。(4)搭建了机械臂硬件平台,并基于该平台对上文中的无模型与基于固定模型的迭代学习控制算法进行了初步实验。实验结果显示,由于观测误差等因素的存在,系统输出存在一定波动;但是随着迭代的进行,机械臂末端执行器的输出轨迹与期望轨迹之间的误差能够逐渐减小。本文对上述实验结果进行了误差分析。
[Abstract]:In the production process of light industry, handicraft industry and other industries, there is a large number of high intensity repetitive work, this kind of work is suitable for the use of robotic arm system instead of human completion. In automobile, electronics and other industries, the mechanical arm has been widely used, and the related control technology of the manipulator system has been relatively mature. However, most of the existing mature manipulator systems use the precision mechanical arm produced by well-known foreign manufacturers. The cost of purchasing and maintaining the system is very high, and it is difficult for the small and medium enterprises in our country to afford the high cost of the equipment. Therefore, it is of great practical significance to select the appropriate low cost manipulator system and put forward the control scheme of the manipulator which is suitable for the actual working conditions, which is of great practical significance for the promotion of small and medium-sized enterprises and the automation degree of the handicraft enterprises. In this paper, a class of manipulator with kinematics controller is selected as the research object, and three kinds of iterative learning controllers are designed to solve the control problems under different conditions. It provides a complete solution for the controller selection of this kind of manipulator system. The main contents of this paper are as follows: (1) A model-free PD type closed-loop iterative learning control algorithm is selected for the unknown manipulator model. Firstly, the iterative learning algorithm is very suitable to solve the trajectory tracking problem of nonlinear systems, so iterative learning control is chosen as the basic control scheme of the manipulator system. In this paper, the characteristics of closed-loop iterative learning algorithm and open-loop algorithm are compared, and a closed-loop iterative learning algorithm is selected, which is effective in suppressing environmental interference. Finally, considering the simple structure of pid controller, Because of the strong robustness and the integral effect of iterative learning process itself on the iterative axis, the PD type closed-loop iterative learning law is selected as the control algorithm for the manipulator system. The simulation results show that the algorithm is effective in solving the trajectory tracking problem of manipulator. (2) an iterative learning control algorithm based on fixed kinematics model is proposed to solve the problem of manipulator reference model. By using the reference model of the manipulator as a priori knowledge and using the reference joint angle obtained from the inverse kinematics solution to guide the iterative learning process, the algorithm can effectively accelerate the convergence rate of the tracking error of the system. The simulation results show that the convergence speed of the system error is improved obviously by using the control algorithm based on the fixed kinematics model. (3) the deviation between the reference model of the manipulator and the actual object is large. An iterative learning control algorithm based on adaptive model is proposed. In this algorithm, the parameters of kinematics model of manipulator are estimated online by Kalman filter, and the reference model of the object is updated after each iteration for the next iterative learning control. The simulation results show that the manipulator system can successfully complete the trajectory tracking task even if the object reference model mismatch is serious by using the adaptive model-based algorithm. (4) the hardware platform of the manipulator is built. On the basis of this platform, the iterative learning control algorithms without model and fixed model are tested. The experimental results show that the system output fluctuates due to observation errors, but with the iteration, the error between the output trajectory of the manipulator end actuator and the desired trajectory can be gradually reduced. In this paper, the error analysis of the above experimental results is carried out.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP241

【参考文献】