基于自适应动态规划的可重构机械臂容错控制方法研究

发布时间：2018-01-05 18:25

本文关键词：基于自适应动态规划的可重构机械臂容错控制方法研究　出处：《长春工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着现代科学技术的不断推进,以及“十三五”国家科技创新规划的部署,针对像可重构机械臂一类具有模块化、灵活性等特点,以及对外界环境与操作任务具有强大适应能力的智能机械设备,将被大力发展应用于深空探测、智能工厂以及核工业等很多应用领域中。然而,当可重构机械臂长期工作在极端的恶劣环境下,其执行机构等元器件不可避免会发生故障。若不及时对其进行处理,将会带来重大的财产损失。因此,研究故障诊断与容错控制方法来解决此类问题以维持系统可靠的性能十分迫切。与此同时,在提高可重构机械臂系统的控制精度,简化控制器结构以及优化能源代价消耗这方面的研究也十分必要,不仅将为复杂机构的机器人以及大型工程机械的实际应用提供有力的技术支持,同时对机械设备向着自动化、智能化以及最优化的方向健康发展具有切实深远的意义。本文首先利用刚体的动力学方程几何公式,通过Newton-Euler迭代算法,建立了可重构机械臂系统动力学模型,并在此基础上,给出了故障系统动力学模型。其次,针对轨迹跟踪问题,为了提高系统的控制精度,优化能源消耗以及简化控制器的结构。结合最优控制理论知识,设计一种基于单网络评价结构策略迭代(Critic-only Policy Iteration,CoPI)算法的可重构机械臂最优跟踪控制方法。利用神经网络构建了单网络评价机构。通过CoPI算法计算得出最优的反馈跟踪控制律,并结合名义控制律共同构成的全局跟踪控制律确保了系统的跟踪误差渐近收敛至零。然后,考虑到可重构机械臂在不同的任务需求中,可能存在模块增减的情况,则分散控制思想更加适合于可重构机械臂控制。同时为了兼备“最优化”思想,故提出一种基于观测器评价结构自适应动态规划(Adaptive dynamic programming,ADP)算法的分散控制方法。为了解决对耦合交联项处理的难点,本章结合替换思想与观测器技术,采用神经网络对替换处理后的子系统动力学模型进行辨识。利用CoPI算法设计局部反馈控制器,采用自适应鲁棒项来抵消系统的整体逼近误差,并结合设计的局部名义控制器共同保证了系统的轨迹跟踪误差渐近收敛至零。最后,针对执行器发生故障的可重构机械臂系统,基于ADP理论和故障观测器,提出一种改进性能指标函数ADP算法的容错控制策略。利用自适应故障观测器对执行器故障的观测值设计一种改进的性能指标函数。将容错控制问题转化为最优控制问题来处理。容错控制器由名义控制器与最优反馈控制器两部分组成,设计名义控制器是为了保证系统的跟踪性能能够稳定到零,最优反馈控制器是为了使跟踪误差以最优的方式收敛于稳定的状态。仿真实验验证了所设计的观测器与控制器的有效性。
[Abstract]:With the continuous advancement of modern science and technology, and the deployment of the 13th Five-Year Plan of National Science and Technology Innovation Plan, it has the characteristics of modularization and flexibility, such as reconfigurable manipulator. Intelligent mechanical equipment with strong adaptability to external environment and operation tasks will be developed and applied in many applications such as deep space exploration, intelligent factories and nuclear industry. When the reconfigurable manipulator works in an extremely bad environment for a long time, its actuator and other components will inevitably fail. If it is not dealt with in time, it will bring great property losses. It is very urgent to study fault diagnosis and fault-tolerant control methods to solve such problems in order to maintain the reliable performance of the system. At the same time, the control accuracy of the reconfigurable manipulator system is improved. It is also necessary to simplify the structure of controller and optimize the cost of energy consumption, which will not only provide powerful technical support for the practical application of robot with complex mechanism and large construction machinery. At the same time, it has a practical and far-reaching significance for the development of mechanical equipment towards the direction of automation, intelligence and optimization. Firstly, the geometric formula of dynamics equation of rigid body is used in this paper. The dynamic model of reconfigurable manipulator system is established by Newton-Euler iterative algorithm, and on this basis, the dynamic model of fault system is given. Secondly, the trajectory tracking problem is addressed. In order to improve the control accuracy, optimize the energy consumption and simplify the structure of the controller, combined with the knowledge of optimal control theory. An iterative critical only Policy Iteration based on a single network evaluation structure strategy is designed. The optimal tracking control method of reconfigurable manipulator is proposed. A single network evaluation mechanism is constructed using neural network. The optimal feedback tracking control law is obtained by CoPI algorithm. The global tracking control law combined with the nominal control law ensures that the tracking error of the system converges to zero asymptotically. Then, considering the reconfigurable manipulator in different task requirements. If there is an increase or decrease of modules, the idea of decentralized control is more suitable for the control of reconfigurable manipulators, and for the sake of "optimization". This paper presents an adaptive dynamic programming based on observer to evaluate the structure of adaptive dynamic programming. In order to solve the difficulty of dealing with the coupling crosslinking term, this chapter combines the idea of substitution with observer technology. The neural network is used to identify the dynamic model of the subsystem after replacement, the local feedback controller is designed by CoPI algorithm, and the adaptive robust term is used to offset the global approximation error of the system. Combined with the designed local nominal controller, the trajectory tracking error of the system converges asymptotically to zero. Finally, for the reconfigurable manipulator system with actuator failure, based on ADP theory and fault observer. An improved fault-tolerant control strategy based on ADP algorithm for performance index function is proposed. An improved performance index function is designed by using the observations of adaptive fault observer for actuator faults. The problem of fault-tolerant control is transformed into the most important problem. The fault-tolerant controller consists of a nominal controller and an optimal feedback controller. Nominal controller is designed to ensure that the tracking performance of the system can be stabilized to zero. The optimal feedback controller is designed to converge the tracking error to a stable state in an optimal manner. The simulation results show the effectiveness of the proposed observer and controller.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP241

【参考文献】