【作者】 熊静；

【导师】 罗志增；

【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】 智能下肢假肢是近些年来机器人学和生物医学工程领域倍受关注的研究方向,下肢姿态检测及运动状态预测是智能下肢假肢控制的基础。研究智能假肢,为截肢者提供性能优良的假肢,有助于提高截肢者的生活质量,对于构建和谐社会具有重要的意义。目前国外已经出现智能化和仿生度很高的智能下肢假肢,而国内在这一领域的研究还存在较大差距,尚未研制出成熟的智能下肢假肢产品。因此从事智能下肢假肢的探索与研究,对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。本论文结合“自适应智能膝上假肢关键技术的研究”课题(863项目2008AA04Z212)的部分研究任务展开,针对单侧膝上下肢截肢的残疾人用智能假肢,获取下肢姿态的多源运动信息,采用有效的预测算法进行下肢运动状态的预测,为进一步实现下肢假肢的自适应控制奠定基础。本文具体完成的研究工作有:(1)针对下肢姿态检测,建立了一个由足底压力、膝关节角度和肌电信号等传感器和测量装置组成的下肢多源运动信息采集系统。具体方法是,运用鞋底嵌入三个PVDF力传感器的方法测取足跟和足趾区域的压力;利用两个ADXL203加速度传感器组合获取膝关节的屈伸角度;运用MyoTrace 400采集仪获取股内侧肌、半腱肌、长收肌、阔筋膜张肌的肌电信号。(2)根据下肢运动的步态周期规律,给出了一种步态和时相结合的下肢运动状态的划分方法,使之能满足智能下肢假肢运动控制的需要。研究分析足底压力、膝关节角度和肌电三种信号的特点,提出了相应的特征提取方法:对下肢表面肌电信号提出了一种基于相关性分析的特征提取方法;对足底压力信号采用阈值化的判断方法;为了减少膝关节角度信号的处理工作量,以同一步态时相中的膝关节角度均值作为特征。(3)运用CPN神经网络技术对下肢运动状态进行分类识别。网络输入为多源运动信息中的肌电信号、足底压力和膝关节角度信号等组成的特征向量,网络输出为待识别的运动状态,并将识别结果与课题组成员所提出的方法进行了比较分析。(4)研究了RBF径向基函数神经网络和ANFIS自适应模糊神经推理系统预测算法,在分析比较两种方法对下肢膝关节角度变化预测的基础上,用ANFIS完成了基于多源运动信息的下肢运动状态预测。实验结果表明,ANFIS能够有效的预测下肢运动状态,预测平均误差能满足下肢运动状态预测精度的需要,为建立下肢运动自适应预测控制系统奠定了基础。论文取得的主要创新点:本文建立了基于多源运动信息的下肢运动状态预测系统,提出将肌电信号、足底压力信号和膝关节角度信号的组合特征输入CPN神经网络识别出当前的运动状态,然后将此刻及以前四个时刻的运动状态参数输入ANFIS自适应模糊神经推理系统预测下肢未来的运动状态。更多还原

【Abstract】 In recent years, intelligent lower limb prosthesis is an attentive research project in the fields of robotics and biomedical engineering, the detection of lower limb’s posture and prediction of motion state is the basis for control of intelligent lower limb prosthesis. To study intelligent prosthesis provides amputees with prosthesis which have good performance to improve their life quality, and has a great significance for constructing harmonious society. Intelligent prosthesis with high capability and bionic degree has been realized overseas, but the domestic research in the field still has a large gap with their study, there is no mature intelligent artificial leg product. The exploration and development of intelligent artificial leg is of great significance to shorten the gap with the developed countries and to promote Rehabilitation Medicine Engineering Technology and prosthesis industry of China.Combining some research tasks of“the research on the key technology of adaptive intelligent knee prosthesis”(the national 863 Project 2008AA04Z212), this paper aim at the disabilities which have only one knee lower limb amputation to use the intelligent prostheses, obtain multi-source motion information of the lower limb posture, adopt effective prediction algorithms to predict the moving state of lower limb, lay the foundation for the study of Adaptive Motion Control. The major research works are as follows:(1) For the detection of lower limb posture, a set of system for collection of lower limb’s multi-source motion information was built, which consisted of some sensors and the measurement device of plantar pressure, knee angle and EMG signal. The detailed method was as follows: The shoe attached PVDF force sensor was used to obtain heel and toe regional pressure; two ADXL203 accelerometers was combined to obtain the angle of knee joint; the MyoTrace 400 was used to obtain EMG signals of the Vastus medialis muscle, the semitendinosus muscle, the long adductor muscle and the tensor fascia lata.(2) According to the law of lower limb movement’s gait cycle, a method combining gait and phase for mode subdivision of lower limb movement was given, which met the need of motion control of the intelligent artificial leg. After researching and analyzing the characteristics of plantar pressure, knee angle and EMG three signals, the corresponding effective feature extraction methods were proposed in this paper: the feature extraction based on correlation analysis was used in EMG; the method of judging based on threshold was used in plantar pressure signal ; in order to reduce the workload of processing the knee angle signal, the mean of the knee angles under the same gait phase was used as the feature of the angle signal of knee joint.(3) The CPN neural network was used to identify the moving state of lower limb. The input of the network was a concatenation feature vector, consisted of the features of EMG signal, plantar pressure and the knee angle signal. Then the results of identification through the CPN network and the methods given by Task Force members were compared and analyzed.(4) The prediction algorithms of Radial Basis Function neural network(RBF) and Adaptive Neural Fuzzy Inference System(ANFIS)were studied. On the basis of comparison and analysis of the forecasting results of knee joint angle changes through the two methods, the prediction of the lower limb moving state based on the multi-source motion information was performed by using ANFIS. The experimental results showed that the moving state can be predicted effectively by using ANFIS and the average prediction error met the need of the prediction accuracy for the lower limb moving state, the foundation for the establishment of the lower limb motion’s adaptive prediction control system was laid.The major innovation is as follow:The prediction system for the lower limb moving state based on the multi-source motion information was established, the idea was like this: the concatenation feature vector consisted of the features of EMG signal, planter pressure and the knee joint’s angle signal was input to the CPN neural network to identify the current motion state of lower limb, Combined with the first three parameters of motion state to predict future motion state of lower limb by using Adaptive Fuzzy Neural Inference System(ANFIS).更多还原