电驱动刚性机械臂神经网络控制新方法

【作者】 罗伟林；

【导师】 林桐；

【作者基本信息】 福州大学 ， 固体力学， 2002， 硕士

【摘要】 本文探讨了具有不确定性因素的机器人系统的跟踪控制问题。分析了机器人系统的动力学特性，建立了系统的运动微分方程。对由机械子系统和电气子系统构成的机器人系统，结合逆运动学补偿方法，综合运用鲁棒控制和神经网络控制技术给出两类控制器的设计原则。 (1)带神经网络辨识器的鲁棒控制器设计。在考虑干扰抑制性能的前提下，对系统建模误差的处理运用基于H_∞控制理论的鲁棒控制策略。对控制器设计中遇到的电流期望值一阶导数的处理运用二层前馈神经网络控制策略。采用了一种新的网络权值动态演化算法。 (2)利用神经网络的非线性映射能力来识别系统的建模误差。整个系统的控制器由两个二层前馈神经网络构成，同时控制器设计时还考虑了系统对干扰的抑制性能。 对以上两种控制方法进行了稳定性分析，证明了跟踪误差和神经网络权值误差的一致终值有界。同时给出了误差上界值和控制器参数值的关系式，对控制器的设计具有指导意义。 同以往的控制方法相比，本文设计的控制器具有结构简单、实时性好、控制精度高和鲁棒性能强的特点，这在机器人实际操作中具有一定的意义。更多还原

【Abstract】 This paper concerns to the tracking control problem of robots with uncertainties. It analyses the dynamical characteristics of the robots and presents the dynamical differential equations of the system. The controller design of the robot system which is made up of mechanical subsystem and electrical subsystem results from the techniques of robust control and neural network control, in addition to the method of anti-dynamics.(1) The design of robust controller with NN. To care about the ability of suppressing disturbance, a robust controller which is underlined to the H control theory is designed to deal with the model errors. A two-layer feed forward NN which involves a novel weights algorithm is used to identified the first-order differentials of the expected current.(2) The former model errors can also be regarded as nonlinear functions, which makes it that they can be identified by NN because of its nonlinear function ability. The overall controller of system is made up of two two-layer feed forward NNs.Stability analysis is made for the above two control algorithms, which results in the uniformly ultimately bounded stability of the tracking errors and weights. What’s more, we present the relationship between the errors upper bound and the controller parameters, which is valuable to the design of the controller.To compare former controllers, the ones designed in this paper are characterized by simple structure, good real time property, fine control precision, and guaranteed robustness, which makes it constructive in the real-time control of robots.更多还原