【作者】 王昕；

【导师】 杨建华；

【作者基本信息】 浙江大学 ， 控制科学与工程， 2019， 硕士

【摘要】 作为康复医学工程和机器人工程学科交叉领域的研究热点,下肢康复机器人是目前前沿且日渐成熟的一个研究方向,由患者主动运动意图参与并主导的下肢康复训练可激发患者中枢神经元的代偿和重组,同时增强下肢康复机器人的高度人机耦合性,更利于患者下肢运动机能的恢复。对人体运动意图定性、定量捕获并快速解析,准确映射到柔顺控制系统输入,这一系列人体交互信息与运动意图之间映射关系的探究具有诸多挑战。本文针对上述难点开展了面向下肢康复机器人的运动意图识别技术研究,重点突破运动模态及步态子相辨识、运动意图预测等关键技术,搭建实验平台,验证所提算法的有效性,进而为基于运动意图推理的下肢康复机器人主动运动控制奠定技术基础。本文的研究内容和贡献如下:(1)构建基于外辅式下肢康复机器人实验系统,包括6层级足底压力感知系统、关节角感知系统和下肢位姿感知系统,完成其硬件平台、软件平台和数据平台的资源分配和功能性配置,设计运动数据感知采集实验完成自建数据集Motion modality&Gait phase的采集、检验和分析。(2)针对运动模态及步态子相辨识的问题,提出了一种组合特征选择算法Filter-BC-MFB-SVM,获得使分类器分离度最大的特征集合,并结合分类任务设计TM-SVM算法,使模型预测阶段效率加快、层级误差累计效应减弱,最终完成静坐、站立和行走三种运动模态和跖屈控制相(CP)、背屈控制相(CD)、跖屈动力相(PP)和摆动相(SW)4个步态子相的准确分类。(3)针对运动意图预测的问题,采用Stacking方法将通过多预测模型ARIMA、SVR、RF和XGboost对关节角的点估计得到的预测值进行融合,利用统计学区间估计方法构建置信区间和预测区间,利用区间约束法则规划出新的预测值,精准、快速地得出判定当前相位下一时刻人体下肢运动轨迹。更多还原

【Abstract】 As a research hotspot in the field of rehabilitation medical engineering and robot engineering,the lower limb rehabilitation robot is a frontier and increasingly mature research direction.The lower limb rehabilitation training which is participated and led by the patient’s active movement intention can stimulate the compensation of the central nervous system of the patient.Recombination,while enhancing the high degree of human-machine coupling of the lower limb rehabilitation robot,in favor of the recovery of the patients with lower limb motor function.The exploration of the mapping relationship between human interaction information and motion intentions has many challenges for qualitative,quantitative capture and rapid analysis of human motion intentions and accurate mapping to compliant control system inputs.In this paper,the research on the motion intent recognition technology of the lower limb rehabilitation robot is carried out for the above difficulties.The key technologies such as motion modality and gait subphase identification and motion intention prediction are broken.The experimental platform is built to verify the effectiveness of the proposed algorithm,which is based on the intention inference of the lower limb rehabilitation robot active motion control technology foundation.The research content and contributions of this paper are as follows:(1)The experimental system based on external and auxiliary lower limb rehabilitation robot is constructed,including 6-level plantar pressure sensing system,joint angle sensing system and lower limb posture sensing system.The resource allocation and functional configuration of the hardware platform,software platform and data platform are completed.The motion data perception acquisition experiment completes the acquisition,inspection and analysis of the self-built dataset Motion modality&Gait phase.(2)Aiming at the problem of motion mode and gait subphase identification,a combined feature selection algorithm Filter-BC-MFB-SVM is proposed to obtain the feature set that maximizes the classifier’s separation degree.The TM-SVM algorithm is designed in combination with the classification task,which improves the efficiency of model prediction stage and weakens the cumulative effect of horizontal error.Finally,the three motion modes of sit-sitting,standing and walking and the accurate classification of four gait sub-phases of CP phase,CD phase,PP phase and SW phase are completed.(3)For the problem of motion intent prediction,the stacking method is used to fuse the predicted values of the joint angle estimation by the multi-prediction models ARIMA,SVR,RF and XGboost.The confidence interval and prediction interval were constructed by statistical interval estimation method.By using the interval constraint rule,the new predicted value is planned,and the trajectory of human lower limb movement at the next moment of the current phase is obtained accurately and quickly.更多还原