【作者】 李庆玲；

【导师】 孙立宁；

【作者基本信息】 哈尔滨工业大学 ， 机械电子工程， 2009， 博士

【摘要】 上肢康复训练机器人是医疗机器人的一个重要分支,它以偏瘫上肢的运动再学习理论为依据,融合了康复医学、人体解剖学、机械学、计算机科学及机器人学等诸多学科技术,是当前国内外研究的热点。康复机器人作为以人体,尤其是以病患个体为作用对象,辅助或替代医师执行患肢康复训练为目的的机器人系统,不同于传统工业机器人,在实现基本功能的条件下,其设计和控制须充分考虑人体安全性、临床操作的可行性、系统功效性、患者可接受性及患者主动参与性等因素。论文在黑龙江省科技攻关重点项目支持下,针对临床康复应用需求,对上肢康复机器人系统设计、上肢康复机器人的表面肌电控制、上肢康复机器人的力控制及偏瘫上肢康复训练策略等方面进行研究,为康复机器人系统设计及康复训练奠定理论及技术基础。论文首先以人体上肢解剖学及运动理论为依据,从临床康复的安全性、有效性、实用性及舒适性角度提出对上肢康复机器人机构、控制、传感器等方面的设计要求。结合临床应用设计适用于不同身高、不同患侧、不同损伤程度的上肢偏瘫患者的新型5DOF外骨骼式上肢康复机器人系统,并详细阐述其机械结构、控制体系、驱动硬件、传感器的设计及运动学分析。表面肌电信号作为神经—肌肉运动产生的电信号富含了肌肉的运动状态信息,因此将表面信号引入康复机器人系统,实现人体上肢运动的辨识,预测人体主动运动意图。首先,基于表面肌电信号的短时平稳假设,提出了表面肌电信号的分段自回归建模方法,并利用主成分分析与多特征融合实现四通道表面肌电信号的特征提取。通过不同样本集训练条件下的单一BP神经网络分类测试实验,以错误率和网络训练迭代次数为指标,证明了提出的特征提取算法较传统自回归模型描述方法更具有优越性,并得到了提出的特征提取算法在非线性特征空间可分性更强的结论。在此基础上,针对单一BP神经网络分类器的识别率不稳定特点,基于Adaboost集成分类思想提出度量输出信息自适应加权的表面肌电动作模式集成分类算法。实验结果表明集成分类器由于综合考虑了多个基分类器间的互补信息,较单一神经网络分类器具有更强的泛化、推广能力,可提高康复机器人系统对人体上肢运动意愿的判断准确率。上肢主动运动对偏瘫患者的康复进程具有较强的推动作用,而关节力矩信号是康复过程中人体上肢主动运动意图的直接体现。基于阻抗控制理论研究基于关节力矩信号的上肢主动康复训练方法。首先建立上肢康复机器人静力学模型,研究关节力矩电压信号的预处理及空载力矩去除方法。然后根据末端力阻尼控制策略和关节刚度控制策略提出“阻尼式”和“弹簧式”两种上肢主动训练策略:阻尼式上肢康复训练将“人—机”作用关系建模为机械阻尼,患者上肢主动运动、康复机器人被动跟随并以速度阻尼形式为人体上肢提供康复所需的运动阻力;“弹簧式”主动抗阻训练中康复机器人以关节弹簧形式为上肢偏瘫患者提供刚性抗阻力,通过设置虚拟弹簧的刚度系数,可以达到患者“拉”动不同刚度的虚拟弹簧的训练效果。针对临床偏瘫患者的康复进程提出进阶交互式上肢康复训练策略。康复早期采用基于表面肌电信号的自主性被动康复训练,训练方式为患者健侧上肢控制患侧上肢,达到双臂镜像协调训练的目的。康复中后期采用基于关节力矩信号的主动康复训练。最后基于5DOF外骨骼式上肢康复机器人平台,以多个受试者进行实验研究,实验结果一方面验证了提出的上肢康复机器人系统能够实现不同的康复训练模式,可满足临床康复需要,另一方面验证了表面肌电控制方法与关节力矩阻抗控制方法的正确性。更多还原

【Abstract】 As an important branch of medical robot, rehabilitation training robot for hemiplegic upper limbs is a hot research topic both at home and abroad. Based on motor relearning programme, it combines many technology fields such as rehabilitation medicine, human anatomy, mechanics, computer science, and robotics, etc. Different from traditional industrial robots, rehabilitation robots work for human bodies, especially for diseased ones, to assist or displace therapists to carry out rehabilitation exercises. Meanwhile, on the basis of basic functions, human safety, feasibility of clinic operating, system efficiency as well as acceptability and active participation of the patients must be taken into consideration while designing and controlling rehabilitation robots. According to clinic application demands, this dissertation makes researches on system design, surface electromyogram control, force control of rehabilitation training robot for hemiplegic upper limbs and rehabilitation training strategies supported by Heilongjiang Key Science and Technology Project, and lies the theoretical and technical foundation for rehabilitation robot design and training.Taking safety, validity, practicability and comfort into consideration, the design rules for mechanism, control and sensor of rehabilitation for upper limbs is proposed on the basis of human upper limb anatomy and motion mechanism. Under the guidance of rules, a novel 5 DOF exoskeletal upper limb rehabilitation robot for patients in various heights, disable side, impaired degree is developed. The mechanical structure, drive hardware, sensor and kinematics of rehabilitation is described in detail.Surface electromyogramm signal contains motion condition information as the electric signal generated with nerve-muscle motion. It is introduced into rehabilitation robot system to recognize upper limb motions for extracting human active motion intention. Combining with Principal component analysis and multi-features fusion, subsection auto-regressive model method for sEMG signal was proposed to extract four channels sEMG features base on short time stationary hypothesis firstly. The classification test experiment of single BP neural network with different training sample set proves its stronger splittability than traditional AR model method in nonlinear feature space. Refer to instability of single BP neural network classifier, adaptive weighted motion mode ensemble classification algorithm according to output information was proposed based on Adaboost ensemble classification concept. Experiments showed that ensemble classifier has greater generalization than single ones by integrating complement information between each single ones. It can improve judging accuracy of upper limb motion intention for rehabilitation robot.Upper limb active motions have great promotion for rehabilitation course of hemiplegic patients. Joint torque is the direct reflection of active motion intention. So, active rehabilitation training method for upper limbs based on joint torque is studied on the basis of impedance control theory. Static model of rehabilitation robot for upper limbs is built and preprocessing method for joint torque voltage and no load torque removal method are discussed. Then, two active training methods-“damping”and“springing”are proposed base of damping control strategy and joint stiffness control strategy respectively. The former modeled the human-machine interaction as mechanical damping. Rehabilitation robot follows patients’active motion and provides needed resistance with velocity damp. In“springing”anti-resistance training, rehabilitation robot provides rigid resistance for upper limbs with pattern of spring. Patients fell like pull different virtual spring by setting rigid coefficient.The stepping and interactive rehabilitation strategy for upper limbs of hemiplegic patients is proposed. sEMG based self passive rehabilitation exercises is for early recovery stage. The disable limb is controlled by healthy one for harmonious mirror image exercises of all two. Active rehabilitation exercises based on torque signal is for middle to late period. At last experiments are carried on with healthy people as trails on the flat of 5DOF exoskeletal rehabilitation robot for upper limbs. The result of experiments verifies that developed rehabilitation robot can satisfy clinical rehabilitation by various rehabilitation exercise modes, and validates the control method based on sEMG and force.更多还原