面向速度适应的动力小腿假肢蹬地时刻在线识别

On-line recognition of push-off for powered transtibial prosthesis towards speed adaptation

本文开展了针对残疾人穿戴动力假肢行走时蹬地时刻的在线识别及面向速度的适应性问题研究.实验中,残疾人被试穿戴我们自主设计的智能动力小腿假肢.该动力假肢通过角度传感器测量踝关节角度,利用角度传感器信号可以将运动划分为两个模式(模式转换时刻即为蹬地时刻).两个惯性测量单元和一套电容传感系统被用来识别这两种运动模式及蹬地时刻.实验中,我们首先采集了被试穿戴动力假肢在0.9 m/s速度下行走的数据进行模型训练,然后基于此模型研究在不同速度下(0.7, 0.9, 1.1 m/s三种情况)对蹬地时刻的在线识别效果.结果如下:首先,在不同的速度下,针对两种模式的在线识别分别取得了97.59%, 98.17%, 93.39%的准确率;其次,我们发现,在三种速度下行走的每一步均可以正确识别出蹬地时刻,识别出的蹬地时刻与准确的蹬地时刻之间存在时延,通过对时延的统计分析,结果表明三种速度下的识别时延统计结果相应为1.5±8.75 ms,–12.38±5.39 ms,–29.52±15.64 ms.结合识别准确率与时延分析,可以得出结论:本文提出的在线模式识别方法识别蹬地时刻,具有准确率高、延时短的特点,同时显示出良好的速度适应性.

The paper investigates the on-line recognition of push-off for an amputee who wears powered prosthesis at different speeds.In the experiment,the subject wears the powered prosthesis designed by our research group.The angular sensor integrated in the prosthesis measures the ankle angle of prosthesis.The moment corresponding to the maximum ankle angle is the moment of push-off.We divide one stride into two modes (mode 1to mode 2)based on ankle angle (the transition moment from mode 1 to mode 2 is the push-off moment).Two inertial measurement units and a capacitance sensing system are used to recognize the two locomotion modes and push-off.We first collect the walking data that the subject walks on treadmill at the speed of 0.9m/s to train model,then carry out the research based on the model under different speeds (0.7,0.9,1.1m/s).Under different speeds,on-line real-time recognition accuracies are 97.59%,98.17% and 93.39%,respectively.Besides,specific push-off in one stride could be recognized with no error. There is time delay between the recognition result of push-off and the real push-off of the powered prosthesis,which appears in advance or behind (corresponding to 1.5±8.75ms,-12.38±5.39ms,-29.52±15.64ms).The results suggest that the method of mode recognition is efficient and feasible to recognize push-off with high accuracy and short delay,which shows good speed adaptability.