【作者】 刘珊；

【导师】 王永骥；

【作者基本信息】 华中科技大学 ， 控制理论与控制工程， 2008， 博士

【摘要】 随着上肢偏瘫患者数量的增多,上肢康复机器人越来越多地引起人们的重视。但现有的上肢康复系统大多采用开环控制系统结构,较少考虑到患者的主动运动意识和反馈信号。本论文针对上肢康复机器人的运动控制问题,先对人体上肢的运动控制机理进行研究,再结合得到的结论,对上肢康复机器人运动控制进行研究。本文的主要工作和贡献包括以下几个方面:研究了矢状面内人体上肢的姿态控制问题。首先分析了矢状面内人体上肢的运动特性,建立矢状面内人体上肢神经肌肉骨骼系统的数学模型,采用多层感知器神经网络实现上肢姿态控制系统,控制系统的持续时间根据反馈误差实时估计得到,比较Levenberg-Marquardt学习算法和弹性BP算法在上肢姿态控制问题中的适用性。研究了矢状面内人体上肢的大幅度姿态运动控制问题。基于小脑神经网络思想,结合模糊控制和滑模控制理论,提出了一种滑模模糊小脑神经网络。该网络利用滑模面函数转换网络输入信号,提出基于量化方式的多维接受域函数实现联想单元与接受域空间之间的映射,有效地降低了网络的存储空间。网络参数的训练分两阶段完成,粗调阶段以滑模控制器训练网络参数,微调阶段以粗调阶段训练结果为基础,进一步加强控制系统的稳定性。针对人体上肢的轨迹跟踪控制问题,提出了一种进化对角回归神经网络,网络的结构由遗传算法与进化计算结合的混合学习算法优化得到,网络的参数由附加动量因子的自适应动态反向传播学习算法训练得到。这种具有动态映射能力和生物进化特性的神经网络更适合于矢状面内人体上肢这类多变量非线性系统的轨迹跟踪控制,可以进一步用于康复机器人控制等问题中。讨论了人体上肢的实时轨迹跟踪控制问题。在滑模模糊小脑神经网络的基础上,提出了一种滑模对角回归小脑神经网络。该网络在联想层节点中引入自回归单元,使网络具有更好的动态映射能力。这种动态小脑神经网络具有更好的跟踪平滑性和更强的鲁棒性,适合于多变量不确定非线性系统的动态控制。研究了上肢康复机器人的姿态控制问题。分析了上肢康复机器人的功能和结构,对上肢康复系统的机械结构部分建立了动力学模型。根据临床观察和患肢的病理分析,设计患者的主动作用函数。根据上肢康复机器人的控制需求,结合人体运动控制思想,采用鲁棒控制实现了上肢康复系统的大幅度姿态运动控制。根据李亚普诺夫稳定性定理,设计鲁棒控制律,并在仿真和实验中验证该控制律的有效性。讨论了上肢康复机器人的轨迹跟踪问题。基于逆向动力学补偿和滑模控制,提出一种多变量滑模控制策略,较好地实现了上肢康复系统的机械结构模型的轨迹跟踪控制。最后,对全文进行了总结,并指出了下一步需要进行的工作。更多还原

【Abstract】 With the increasing number of the hemiplegia patient, persons regard the rehabilitation therapy of the arm movement functional disorder as a more important thing. But, the current rehabilitation systems are driven by the motor, and have an open-loop control structure. The active movement consciousness of the patients and the feedback signal during the therapy are not usually been considered. This dissertation is to study the human arm motion control in the sagittal plane and the rehabilitation robot control based on the research results of the human arm movement mechanism.The research work has been done in this dissertation includes the following aspects:The posture control of the human arm in the sagittal plane is discussed. The dynamic characteristic of the human arm in the sagittal plane is analyzed, and then the arm is modeled as a neuro-musculoskeletal model with two degrees of freedom and six muscles. A multilayer perceptron network containing feedforward and feedback control modes is used. The duration of the movement is regulated according as the current feedback states. The Levenberge-Marguardt training algorithm and the resilient back propagation algorithm are compared.Based on cerebellar model articulation controller, fuzzy control, sliding mode control scheme, and the dynamic characteristic and the control requirement of the human arm, a sliding-mode-based fuzzy cerebellar model articulation controller is investigated. The sliding mode function is used to transfer the input variables; a quantization mode based multidimensional receptive field function is presented. The network is trained at two learning stages to guarantee the stability of the control scheme.An evolutionary diagonal recurrent neural network is presented for trajectory tracking control of the human arm in the sagittal plane, in which hybrid genetic algorithm and evolutionary program strategy is applied to optimize the network architecture and an adaptive dynamic back propagation algorithm with momentum is used to obtain the network weights, Lyapunov theory can be implemented to guarantee the convergence of the control system. This dynamical network is more suitable for trajectory tracking control of a multi input multi output nonlinear system, and can be further applied into the rehabilitation robot control.Trajectory tracking control in real time of the human arm in the sagittal plane is further discussed. A sliding-mode-based diagonal recurrent cerebellar model articulation controller is presented, in which recurrent units are introduced in the association layer to add the dynamic mapping ability of the network. Compared with other control methods, this network has a better dynamic and static performance and better robustness.The posture control of the rehabilitation robot for the human arm is investigated. The function and the structure of the Robotic Assisted Upper Extremity Repetitive Therapy (RUPERT) are analyzed, and then the forward simMechanics model, the mathematic model, and an interactive robust controller are presented. The functions of patient’s active action are designed depending on the motion ability of their arms. The performance of the controller is tested by simulations and experiments.A multi input multi output sliding mode control scheme is designed for the trajectory tracking control of the multi-joint link robot, the structural model of the RUPERT-hemiplegia arm system.At last, this dissertation is summarized, and the further work is discussed.更多还原