多自由度肌电假肢的比例同步控制研究

发布时间：2018-05-16 03:34

本文选题：肌电信号 + 连续估计　；参考：《杭州电子科技大学》2017年硕士论文

【摘要】：表面肌电信号(Surface Electromyography,sEMG)是肌肉收缩时所产生的动作电位在皮肤表面叠加而成,与肢体的运动直接相关,可以用来反应人的运动意图。由于sEMG具有蕴含信息丰富、采集方便等特点,在运动控制、康复医疗等领域具有重要应用价值。本文以肌电假肢多自由度比例同步控制为立足点,设计实验采集上肢关节肌肉动作所产生的肌电信号和运动参数,结合肌肉协同理论从中枢神经系统协同控制人体运动的角度出发,着重分析协同元、激活系数和角度之间的关系,并应用肌肉协同的模型解决肌电假肢的比例同步控制问题。首先,相比模式分类控制方法,连续估计控制方法更符合人体运动规律,可以类人实现关节的比例同步控制。本文结合人体前臂常用动作模式,设计了上肢三个自由度的独立和组合动作实验,通过运用三维运动捕捉系统同步采集前臂肌电信号和对应动作的关节角度,为后续研究打下数据基础。其次,引入肌肉协同理论来进一步解释和解决人体运动过程中多自由度运动协调问题。通过分而治之的方法确定对应前臂六个独立动作的六个协同元,利用非负矩阵分解(Non-negative Matrix Factorization,NMF)算法对肌电信号的均方根特征分解提取出表示中枢神经系统低维控制参数的肌肉协同元;利用非负最小二乘算法提取出表示对协同元刺激程度的激活系数。通过对比激活系数和关节真实运动轨迹,发现激活系数可以反映出动作类别和肌肉激活程度,但是激活系数并不能完全反映出上肢的实际运动轨迹,动作之间的激活系数也没有充分解耦。然后,为了进一步解决激活系数和运动轨迹之间的偏差问题,分解各个动作激活系数之间的耦合性。通过人工蜂群算法优化的支持向量回归(Support Vector Regression,SVR)算法和BP神经网络回归算法分别构建了映射激活系数到关节角度的SVR激活模型和BP激活模型,利用建立的激活模型从采集的表面肌电信号得到关节运动的连续估计。对两个关节独立和组合运动的估计实验表明,SVR和BP激活模型均能获得较好的估计精度,但是SVR激活模型可以获得更稳定和更精确的估计效果。接着,为了分析关节动作的快慢程度对SVR激活模型估计效果的影响,设计了掌关节和腕关节三个自由度动作分别在高速、中速、低速运动模式下的动作实验。根据相关系数、均方根误差和t检验分析估计结果,指出SVR激活模型在高速和中速状态下可以获得良好的估计效果。最后,设计了肌电假肢比例同步控制的在线仿真实验。通过独立和组合动作的在线控制任务,测试SVR激活模型对单自由度和多自由度动作的比例同步控制效果。本文研究成果给肌电假肢的控制方法提供了一种新的方案。
[Abstract]:Surface electromyography (EMG) is a superposition of action potential produced by muscle contraction on the skin surface, which is directly related to the movement of the limbs and can be used to reflect the intention of human motion. Because of its rich information and convenient collection, sEMG has important application value in the field of movement control, rehabilitation and medical treatment. In this paper, the multi-degree-of-freedom proportional synchronous control of EMG prosthesis is taken as the foothold, and the EMG signals and motion parameters produced by the muscle movement of the upper limb joint are collected experimentally. Based on the theory of muscle coordination, the relationship among synergetic elements, activation coefficients and angles is analyzed from the point of view of central nervous system coordinated control of human body movement. The model of muscle coordination is applied to solve the problem of proportional synchronous control of myoelectric prosthesis. Firstly, compared with the pattern classification control method, the continuous estimation control method is more consistent with the human motion law, and it can realize the proportional synchronous control of joints. In this paper, combined with the common action mode of human forearm, the experiment of three degrees of freedom of upper limb is designed. The electromyography signal of forearm and the joint angle of corresponding action are collected synchronously by using three-dimensional motion capture system. To lay the data foundation for the follow-up research. Secondly, the theory of muscle coordination is introduced to further explain and solve the problem of multi-degree-of-freedom motion coordination. Using the divide-and-conquer method to determine the six synergetic elements corresponding to the six independent actions of the forearm, Non-negative Matrix factorization (NMF) algorithm is used to extract muscle synergists representing the low dimensional control parameters of central nervous system (CNS). The non-negative least squares algorithm is used to extract the activation coefficients representing the stimulus degree of the synergists. By comparing the activation coefficient with the real motion trajectory of the joint, it was found that the activation coefficient could reflect the type of movement and the degree of muscle activation, but the activation coefficient could not completely reflect the actual movement track of the upper limb. The activation coefficients between actions are also not fully decoupled. Then, in order to solve the problem of the deviation between the activation coefficient and the motion trajectory, the coupling between the activation coefficients of each action is decomposed. The SVR activation model and BP activation model of mapping activation coefficient to joint angle are constructed by the support vector regression support Vector regression algorithm and BP neural network regression algorithm optimized by artificial bee colony algorithm, respectively. The continuous estimation of joint motion is obtained from the collected surface EMG signals using the established activation model. The experiments of joint independent and combined motion estimation show that both SVR and BP activation model can obtain better estimation accuracy, but SVR activation model can obtain more stable and accurate estimation results. Then, in order to analyze the influence of the speed and slowness of joint action on the estimation effect of SVR activation model, three degrees of freedom (DOF) motions of metacarpal joint and wrist joint were designed in high speed, middle speed and low speed motion mode respectively. According to the correlation coefficient, root mean square error and t test, it is pointed out that the SVR activation model can obtain good estimation results at high and medium speed. Finally, the online simulation experiment of proportion synchronous control of myoelectric prosthesis is designed. Through the on-line control task of independent and combined actions, the proportional synchronization control effect of SVR activation model for single and multi-degree-of-freedom actions is tested. The results of this paper provide a new scheme for the control of myoelectric prosthesis.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R318.17;TN911.7

【参考文献】