肌电信号选择对下肢关节连续运动估计的影响

介绍了利用肌电信号实现人体运动识别的发展状况和相关方法。针对肌肉选择对下肢关节角度预测性能的影响这一问题,构建了肌电信号组合集,并建立了高阶多项式模型以及提出的改进模型实现对肌电信号子集到关节角度的映射。利用最小二乘法对模型参数进行了辨识,并对比验证了这两种模型的有效性,确认了最佳肌电信号组合与模型自由度。结果表明,提出的改进高阶多项式模型对不同肌电信号组合实现关节角度预测具有更稳健的效果,同时,下肢关节连续运动识别的最佳肌肉选择不一定与关节运动直接相关。     

基于双谱分析的表面肌电信号模式识别

本文提出一种新的用于表面肌电信号分类的方法。这种方法将双谱分析技术应用于表面肌电信号分类来对六种简单的动作进行分类,包括内翻,外翻,握拳,展拳,上切和下切六种动作模式的识别。在以往的表面肌电信号分类中,人们都假设信号满足高斯分布和线性,并且为平稳信号。但是实际的表面肌电信号往往不能满足上面的假设,根据前人对表面肌电信号的研究我们知道,当肌肉收缩低于最大自发收缩的25％时,表面肌电信号所表现的非高斯性是显著的。因此为了获得更多的表面肌电信号的信息和获取更好的表面肌电信号分类的识别率,我们利用双谱分析和主元分析相结合方法对肌电信号进行了分类研究。     

一种肌电假手的自适应模糊控制方法

针对肌电假手的力控制问题,提出了一种基于肌电信号自适应学习的动作识别方法,同时结合模糊神经网络PID控制算法实现肌电信号对假手的控制.设计的肌电信号自适应学习动作识别方法在时域内进行,减小运算复杂度和计算量的同时保证了动作识别精度.采用模糊神经网络PID算法设计了假手握力控制器,在没有位置传感器的情况下保证了假手握力的控制精度.进行了肌电信号动作识别跟踪实验、假手握力跟踪实验、肌电信号控制假手抓取实验,实验结果证明了肌电信号自适应学习动作识别方法和模糊神经网络PID握力控制方法在肌电假手控制中的有效性.     

基于改进能量核的下肢表面肌电信号特征提取方法

由于表面肌电信号(sEMG)具有非平稳、非周期和混沌等特性,导致传统的特征值提取方法存在实时性与准确性难以兼容的问题,基于此提出一种基于sEMG的改进能量核特征提取方法,对采集到的肌电信号进行处理。首先,在EMG振子模型的基础上,详细描述了新提出的"阈值矩阵计数"(threshold matrix count,TMC)特征提取方法;然后,通过在腿部10块不同肌肉表面粘贴肌电传感器来检测下肢不同运动过程中的肌电信号;采集到所需肌电信号后,对10块肌肉上的肌电信号分别进行特征提取,得到10个不同的特征向量x_k,并对其进行分析,选取了4块肌肉作为有效肌肉;最后将有效肌肉的特征向量x_k组合整理,得到特征矩阵X_k,将其输入BP神经网络进行训练,对4种运动模式进行识别。实验结果表明,提出的能量核特征提取方法相比于传统的两种能量核特征提取方法,运算效率分别提升了13倍和9倍;同时,相比常用的时、频域特征提取方法,训练后得到的模型具有更好的稳定性,平均识别精度为95.2%。     

基于表面肌电信号的人体步态事件快速识别方法

针对下肢康复训练机器人主动训练阶段患者运动检测实时性、准确性的需求,提出一种基于动态表面肌电信号的人体步态事件快速识别方法。首先,通过表面肌电信号产生过程数学建模及步态过程中肌肉活动规律分析,给出了基于表面肌电信号强度及其变化特征的步态事件感知原理;其次,以双腿股外侧肌动态表面肌电信号强度及其变化为特征,构建了用于识别支撑和摆动两个步态事件的自适应模糊神经网络模型。实验结果表明：该方法识别结果正确率达95.3%,对足跟触地和脚尖离地事件发生时刻进行识别的平均时间误差分别为21.4ms和24.5ms,同时证明,该方法对步态之间表面肌电信号的差异具有较强的鲁棒性。     

基于HMM的下肢表面肌电信号模式识别的研究

采用隐马尔科夫(HMM)的信号分类识别方法,用以实现下肢步态中各细分动作的更好识别和描述。首先将待分类的肌电信号进行预处理,按等时间间隔对每通道的肌电信号划分不同的段,采用小波变换方法对每个数据段内肌电信号进行多尺度分解,并提取小波分解系数的奇异值构成观察值序列。将步态周期的四种运动模式与HMM状态一一对应,用Baum-Welch算法对HMM参数进行重估训练,再用标定技术处理的前向后向算法和Viterbi算法进行识别,得到的平均识别率高于90%。然后在四种运动模式中分别提取小波特征组成特征编码,送入概率神经网络(PNN)中进行识别。证明HMM的识别效果优于概率神经网络,能够更好地应用于假肢的控制研究。     

采用蝙蝠算法进化相关向量机的假肢步态识别

为了提高假肢穿戴者步态识别准确率,提出了一种高阶过零分析技术分析表面肌电信号的假肢步态识别方法。该方法针对假肢穿戴者步态识别过程中的多分类问题,选择表面肌电信号（surface electromyogram signal,简称sEMG）作为步态识别信息源,将表征时间序列特性的高阶过零分析（higher order zero crossing analysis,简称HOC）方法运用于不同步态下的肌电信号的特征提取,结合相关向量机（relevance vector machine,简称RVM）建立了多分类步态识别模型,然后采用蝙蝠算法（bat algorithm,简称BA）对RVM分类器的核函数参数进行优化。实验结果表明,所提方法与粒子群算法优化相关向量机（particle swarm optimization-relevance vector machines,简称PSO-RVM）及RVM等方法相比,对于平地行走、上楼、下楼、上坡和下坡5种步态的识别准确率均高于PSO-RVM和RVM等方法。     

融合脑电-肌电信号的人机协作装配意图识别方法

针对人机协作装配场景下基于单源生理电信号识别协作意图准确率不高、稳定性不好的问题,首先采用支持向量机方法分别从脑电信号和肌电信号识别单源协作装配意图,然后采用D-S证据理论对多源协作装配意图识别结果进行融合,提出了一种融合EEG-EMG信号的人机协作装配意图识别方法。实验结果表明,所提方法可以有效提高人机协作装配意图识别的准确率和稳定性。     

一种下肢运动意图识别算法性能实时测评系统* .txt

摘要：对腿部运动意图识别算法的实时性能进行综合可靠的评价是实现下肢假肢灵活稳定控制的前提。提出一种逐层分级的基于下肢运动意图识别算法的实时测评方法,对算法的可靠性、稳定性以及运动意图识别速度进行综合评价。利用开发的运动意图识别算法评测系统,对基于肌电信号源和机械信号源的两种运动意图识别算法进行了实时性能测试,发现肌电信号源的算法识别时间大于机械信号源算法,但是其算法稳定性优于机械信号源算法。进一步地,还利用该评测系统有效地区分出正常识别策略与异常识别策略,发现正常策略的动作识别稳定系数比异常策略高25%左右。因此,所提的基于下肢运动意图识别算法的实时测评方法,能够对不同信号源算法以及不同识别策略的实时性能进行有效评价,为智能下肢假肢控制系统开发提供可供参考的测试平台。 .txt     

一种下肢运动意图识别算法性能实时测评系统

对腿部运动意图识别算法的实时性能进行综合可靠的评价是实现下肢假肢灵活稳定控制的前提。提出一种逐层分级的基于下肢运动意图识别算法的实时测评方法,对算法的可靠性、稳定性以及运动意图识别速度进行综合评价。利用开发的运动意图识别算法评测系统,对基于肌电信号源和机械信号源的两种运动意图识别算法进行了实时性能测试,发现肌电信号源的算法识别时间大于机械信号源算法,但是其算法稳定性优于机械信号源算法。进一步地,还利用该评测系统有效地区分出正常识别策略与异常识别策略,发现正常策略的动作识别稳定系数比异常策略高25%左右。因此,所提的基于下肢运动意图识别算法的实时测评方法,能够对不同信号源算法以及不同识别策略的实时性能进行有效评价,为智能下肢假肢控制系统开发提供可供参考的测试平台。     

假肢手恒定力量输出的诱发肌电反馈控制研究

肌电假肢手的连续动态控制有益于提高假肢手的使用性能和用户对假肢手的交互控制能力,产生持续稳定的力量是假肢手需要完成的一种常见任务,但是疲劳、注意力分散等诸多因素都可能导致肌电信号发生变化从而影响假肢手稳定力量输出。本文提出了一种基于电刺激诱发肌电的假肢手稳定力量输出控制方法,通过电刺激尺神经诱发肌电信号对假肢手恒定力量输出进行持续控制。为验证方法可行性,招募了7名健康被试,首先测试了稳定抓握过程中注意力分散对屈肌肌电活动水平的影响,并确定电刺激强度与诱发肌电(M波)幅度的关系模型;在此基础上,测试了电刺激在注意力分散过程中是否能诱发出预期的肌电信号。结果表明注意力分散会引起肌电信号幅度下降,通过电刺激方法可以得到恒定的肌电输出;同时,通过调节刺激参数可以控制诱发肌电强度,可为肌电假肢手的稳定力量输出提供有效的控制信号。     

基于预抓取模式识别的假手肌电控制方法

为解决多自由度假手肌电控制难题,建立一种人手预抓取模式的在线识别方法,并将其应用至HIT-DLR假手的抓取控制。基于Teager-Kaiser能量算子增幅肌电信号在肌肉动作发起时的变化,引入后处理解决噪声影响,提出一种预抓取发起的在线检测方法。针对人手4种预抓取模式,讨论不同肌电信号分段方法,不同时域特征、频域特征和时频域特征以及多种分类方法所能获得的识别成功率。最终建立了基于波形长度及支持矢量机的最优识别方法,成功率可达95%,延迟小于300 ms。肌电控制试验表明,假手可以准确快速地抓取各种不同形状的物体。     

Application of improved morphological filter to the extraction of impulsive attenuation signals

Rotating machinery response is often characterized by the presence of periodic impulses modulated by high-frequency harmonic components. It can be defined with three parameters, which are natural frequency, fault frequency and decay coefficient. In this paper, we propose an improved morphological filter for feature extraction of the above signals in the time domain. Firstly, an average weighted combination of open-closing and close-opening morphological operator, which eliminates statistical deflection of amplitude, is utilized to extract impulsive component from the original signal. Then, according to the geometric characteristic of impulsive attenuation component, the structure element is constructed with an impulsive attenuation function, and a new criterion is put forward to optimize the structure element. The proposed method is evaluated by simulated impulsive attenuation signals with different natural frequencies and vibration signals measured on defective bearings with outer race fault and inner race fault, respectively. Results show that the background noise can be fully restrained and the entire impulsive attenuation signal is well extracted, which demonstrates that the method is an efficient tool to extract impulsive attenuation component from mechanical signals.     

基于DEMD和对称差分能量算子解调的滚动轴承故障诊断

针对机械故障振动信号多为调制信号的特点,为了更好地提取多分量调幅调频信号的幅值和频率信息,提出了基于微分的经验模式分解(DEMD)与对称差分能量算子相结合的解调方法。利用DEMD算法将原始振动信号进行分解,得到若干个单分量信号;对每一个单分量信号进行三点对称差分能量算子解调,得到各单分量信号的瞬时幅值和瞬时频率,并计算出包络谱。将该方法应用于仿真信号和滚动轴承故障信号的诊断,实验结果表明,该方法能有效地提取机械故障信号的故障特征,实现旋转机械故障诊断。     

基于长短时记忆和卷积神经网络的手势肌电识别研究

用表面肌电进行手势识别具有细节信息可选择性和抗外界干扰能力强的优势,但现有方法的适应性和识别准确性还不足.通过在卷积神经网络的基础上增加长短时记忆网络处理层,构筑手势识别模型,它能捕获手势动作过程的肌电时序特征,一定程度上减少了过拟合的现象.利用手势肌电丰富的时频域信息,提取手势肌电的小波包特征图像,并与手势肌电图像一...     

基于振动信号EMD和ICA的刀具破损识别

针对多齿铣削过程中振动信号的特点,提出了一种基于经验模态分解(EMD)和独立分量分析(ICA)相结合的方法,对混叠在振动信号中的铣刀破损信号进行分离。对振动信号进行经验模态分解提取出信号中的所有本征模函数,然后应用fastICA对所提取出的本征模函数进行独立分量分析。利用该方法对铣削加速度振动数据进行了分析,试验表明,该方法可以提取出混合信号中与刀具破损状态相关的故障特征频率成分。     

Fault feature extraction of a rotor system based on local mean decomposition and Teager energy kurtosis

Feature extraction is the most important step for machine fault diagnosis, but useful features are very difficult to extract from the vibration signals, especially for intelligent fault diagnosis based on data-driven technique. An integral method for fault feature extraction based on local mean decomposition (LMD) and Teager energy kurtosis (TEK) is proposed in this paper. The raw vibration signals are first processed via LMD to produce a group of product functions (PFs). Then, the Teager energies are computed using the derived PFs. Subsequently, each Teager energy data set is directly used to calculate the corresponding TEK. A vibration experiment was performed on a rotor-bearing rig with rub-impact fault to validate the proposed method. The experimental results show that the proposed method can extract different TEKs from the mechanical vibration signals under two different operating conditions. These TEKs can be employed to identify the normal and rub-impact fault conditions and construct a numerical-valued machine fault decision table, which proves that the proposed method is suitable for fault feature extraction of the rotor-bearing system.     

Comparison of myocontrol and force control based on fitts’ law model

In recent years, many types of input modalities have been developed and used for a great variety of new devices and machines. To enhance the performance of the human-machine systems, well-designed human-machine interface (HMI’s) between the user and the machine are essential. Biosignal-based HMI’s have been appearing as an alternative to physical HMI’s that have been conventionally used. As a type of biosignal control, the electromyography (EMG) has been investigated as an input modality for prostheses, computers, and robotic exoskeletons. In this study, myocontrol is analyzed through direct and numerical comparison with force control. Mycontrol and force control of visual pointing tasks were tested with EMG and force signals provided as visual feedback, and the controllability of each control mode was evaluated based on Fitts’ law paradigm, which is a general estimation method of speed and accuracy of various movements. The experimental results show that both myocontrol and force control can be modeled using Fitts’ law, even when different types of signals were provided as visual feedback. Among the control modes, myocontrol and force control showed high controllability when force signal was used as visual feedback.     

Determination of muscle fatigue using dynamically embedded signals

This paper explores a new method of analysing muscle fatigue within the muscles predominantly used during microsurgery. The captured electromyographic (EMG) data retrieved from these muscles are analysed for any defining patterns relating to muscle fatigue. The analysis consists of dynamically embedding the EMG signals from a single muscle channel into an embedded matrix. The muscle fatigue is determined by defining its entropy characterized by the singular values of the dynamically embedded (DE) matrix. The paper compares this new method with the traditional method of using mean frequency shifts in the EMG signal's power spectral density. Linear regressions are fitted to the results from both methods, and the coefficients of variation of both their slope and point of intercept are determined. It is shown that the complexity method is slightly more robust in that the coefficient of variation for the DE method has lower variability than the conventional method of mean frequency analysis.     

A METHOD OF HUMAN BIODYNAMIC MEASUREMENT BASED ON VIRTUAL MARKER TECHNOLOGY AND WAVELET ALGORITHM

A new method is reported for human biodynamic measurement, which includes an improved motion measurement system via virtual marker technology and a wavelet algorithm to extract muscle activities from electromyography signals in vivo. Four channels of electromyography signals and twelve virtual markers motion data were collected when the subject did a standard gait experiment. Then, a wavelet algorithm was used to process raw electromyography signals and to extract muscle activity features. The results demonstrated that the proposed method could shorten the motion measurement experiment time and improve the measurement accuracy. The proposed wavelet algorithm could efficiently extract muscle activity features from electromyography signals. The reported method may be potentially used for the human biodynamic research and clinical biomechanics. Possible further improvements for this method are also discussed.