【作者】 徐文良；

【导师】 叶明； 马玉良；

【作者基本信息】 杭州电子科技大学 ， 控制理论与控制工程， 2009， 硕士

【摘要】 研制智能下肢假肢目的是为了改善残疾人生活质量及促进医疗福利事业的发展,同时智能下肢假肢也是近年来机器人学与生物医学工程领域广受关注的研究方向。目前国外已经出现智能化及其仿生度较高的智能下肢假肢,而国内在这一领域的研究状况则不容乐观,还没有成熟的智能下肢假肢产品出现。对智能下肢假肢的探索与研发,为肢体残疾人提供性能优良、价格低廉的智能假肢产品,对于缩短与发达国家的差距,促进我国康复医学工程技术和假肢产业的发展具有重要的意义。智能下肢假肢的成功研制必须以下肢运动姿态的精确识别为前提,考虑到目前下肢运动姿态识别研究中遇到的单一膝关节角度无法辨识复杂多运动状态难题和因残肢肌肉缺损、肌肉疲劳、电极位置改变等因素带来的肌电信号干扰问题,有必要开发出一种获取多种运动力学信息的平台,以及寻找更有效的信号特征分析方法、下肢多运动模式的识别和下肢假肢的自适应控制方法。本文紧扣国家自然科学基金资助项目“膝上假肢的运动力学信息获取与多运动模式控制方法研究(60705010)”的主题,主要完成了以下的研究工作:(1)为获得足够的下肢运动信息,参考下肢运动的特点,并利用课题组现有设备,成功搭建了人体下肢多源运动信息获取系统,为智能下肢假肢的研究奠定了基础。该多源运动信息获取系统包含三个部分:MyoTrace 400肌电信号采集仪(获取大腿不同区域四块肌电的肌肉信号);附着PVDF力传感器的鞋(获取足跟和足趾区域的压力);MTx姿态仪(获取膝关节的屈伸角度)。(2)通过实验方法获取大量多运动模式下的下肢表面肌电信号、足底压力信号和膝关节屈伸角度信号,并采用不同方法对各种信号进行了特征分析。本文对下肢表面肌电信号采用希尔伯特-黄变换的特征提取方法,对足底压力信号提出了一种积分电压比值法的特征提取方法,对膝关节屈伸角度信号提出了角度均值比的特征提取方法。(3)在对下肢表面肌电信号的分析处理过程中,本文首次采用希尔伯特-黄变换(HHT)的方法做了两个方面的探索:提出基于经验模态分解(EMD)的阈值消噪方法。该方法是基于信号和噪声经过EMD后在不同固有模态函数(IMF)上具有的不同特性,即首先对信号进行经验模态分解,然后对高频的IMF分别选用不同的滤波阈值,进行自适应滤波处理。提出HHT边际谱特征提取方法。该方法是基于各层IMF的频率有效程度来选择合适的IMF,通过自适应边际谱的分段方法确定边际谱分段,求取边际谱各段的能量来获得边际谱形状特征。经由实验得到的一种优化的归一化处理方法,最终得到的HHT边际谱的形状特征。(4)在多运动模式识别研究上,本文运用两种比较成熟的神经网络技术(基于L-M改进算法的BP与LVQ)对人体下肢运动姿态进行融合识别。实验结果表明:经过训练的两种网络分类器能正确识别下肢运动模式,基于L-M改进算法的BP网络识别正确率为75%,LVQ网络识别正确率为84%。对实验结果进行分析,发现LVQ具有识别准确率更高、重复性更好及网络训练更加稳定的优点。(5)因为智能下肢假肢运动是复杂多变量、非线性且时变的过程,因此本文选用基于模型参考的神经网络自适应控制技术来对下肢运动进行仿真研究。仿真结果表明,该控制方法用于下肢假肢的控制取得了令人满意的效果。更多还原

【Abstract】 To improve the living quality and welfare benefits of amputees, research has been made on the intellective artificial leg, which is also an attentive research project in the fields of robotics and biomedical engineering in these years. The high-level intelligentized bionic artificial legs have appeared in foreign countries, but research status in this field is not optimistic in China, there is no mature intellective artificial leg product. The exploration and development of intellective artificial leg is of great significance to provide high-performance, low-cost products, and it is also useful to shorten the gap with the developed countries and to promote rehabilitation engineering and prosthesis industry of China.The success of the development of intellective artificial leg must be based on the accurate identification of moving posture, Considering the problems that single angle of knee joint is not enough for complex posture identification and the noise brought by the result of muscle defect, muscle fatigue, changes in the electrode location and other factors disturbs the EMG.. It is necessary to develop a terrace of getting multi-source kinetic mechanical information, seek some good methods of multi-locomotion mode recognition of lower limb and the adaptive control of lower limb prosthesis.This article closely around the National Natural Science Foundation on acquisition of kinetic mechanical information and control method of multi-motion model of AK prosthesis(60705010). In this paper the major research work are as follows:(1) In order to obtain enough kinetic mechanical information, referring to the kinetic characteristics of the lower limb, the paper presents a set of multi-source kinetic mechanical information system by using the existing equipments which create essential conditions for intelligent control of lower limb prosthesis. This information system includes three parts: MyoTrace 400 (obtaining EMG of four different femoral muscles); PVDF force sensor attached to the shoe (obtaining heel and toe regional pressure); MTx sensor (obtaining the angle of knee joint).(2) Through the experimental method, this paper gets much information about SEMG of four different femoral muscles, heel and toe regional pressure and the angle of knee joint, in this paper three feature extraction methods are proposed, the feature extraction based on HHT is used in SEMG, the feature extraction based on the ratio of integral voltage is used in plantar pressure signal and the feature extraction based on the ratio of angle meen is used in the angle signal of knee joint.(3) In the analysis of SEMG, this paper contains two aspects’work through using HHT:Bringing a method of threshold denoising based on empirical mode decomposition (EMD). This method based on different characteristics of signal with noise in different intrinsic mode function (IMFs), fistly the signal can be divided through EMD, finally the high-frequency IMFs are processed by adaptive filter through different threshold;Bringing a feature extraction method based on HHT margin spectrum. This method contains three steps: fistly, the useful IMFs are selected and the self-adapting subsection for HHT margin spectrum is determined; secondly, the energy of each margin spectrum subsection is considered as the sharp feature of the margin spectrum for each SEMG; finally, each dimension of the feature vector is normalized by the method designed based on the experiment.(4) In the research of multi-locomotion mode recognition of lower limb, the two mature neural networks (BP improved algorithm based on L-M & LVQ) are used in identification of moving posture. The experimental results show that the two methods are both suitable to recognize multi-locomotion mode of lower limb, the correct rate of BP neural networks based on L-M is 75%, and the correct rate of LVQ is 84%. Through the results, it shows that LVQ is more recognize rate, more repeatability and more robust.(5) Considering the intellective artificial leg is a complicated system which is nonlinear, multilateral variable and parameters changed with time. In this paper, the neural network based model reference adaptive control technology to simulate research of the lower limb. Experimental result shows that the neural network based model reference adaptive control system is satisfied.更多还原