Zhang Leiyu, Jiao Zhenxing, Li Yongzhen, Chang Yawei
Beijing Key Laboratory of Advanced Manufacturing Technology, Beijing University of Technology, Beijing 100124, China.
Institute for Smart Ageing, Beijing Academy of Science and Technology, Beijing 100089, China.
Biosensors (Basel). 2025 Apr 17;15(4):259. doi: 10.3390/bios15040259.
To enhance wrist impairment rehabilitation efficiency, self-rehabilitation training using healthy-side forearm sEMG was introduced, improving patient engagement and proprioception. A sEMG-based movement recognition and muscle force estimation algorithm was proposed to transmit the estimated results to a wrist rehabilitation robot. Dominant eigenvalues of raw forearm EMG signals were selected to construct a movement recognition model that included a BPNN, a voting decision, and an intensified algorithm. An experimental platform for muscle force estimation was established to measure sEMG under various loads. The linear fitting was performed between mean absolute values (s) and external loads to derive static muscle force estimation models. A dynamic muscle force estimation model was established through linear fitting average s. Volunteers wore EMG sensors and performed six typical movements to complete the verification experiment. The average accuracy of only BPNN was 90.7%, and after the addition of the voting decision and intensified algorithm, it was improved to 98.7%. In the resistance training, the measured and estimated muscle forces exhibited similar trends, with RMSE of 4.2 N for flexion/extension and 5.8 N for ulnar/radial deviation. Under two different speeds and loads, the theoretical and estimated values of dynamic muscle forces showed similar trends with almost no phase difference, and the estimation accuracy was better during flexion movements compared to radial deviations. The proposed algorithms had strong versatility and practicality, aiming to realize the self-rehabilitation trainings of patients.
为提高手腕损伤康复效率，引入了利用健侧前臂表面肌电信号的自我康复训练，提高了患者的参与度和本体感觉。提出了一种基于表面肌电信号的运动识别和肌肉力量估计算法，并将估计结果传输到手腕康复机器人。选择前臂原始肌电信号的主导特征值来构建运动识别模型，该模型包括一个反向传播神经网络（BPNN）、一个投票决策和一个强化算法。建立了肌肉力量估计实验平台，以测量不同负荷下的表面肌电信号。对平均绝对值（s）和外部负荷进行线性拟合，以推导静态肌肉力量估计模型。通过对平均s进行线性拟合建立了动态肌肉力量估计模型。志愿者佩戴肌电传感器并进行六种典型运动以完成验证实验。仅BPNN的平均准确率为90.7%，加入投票决策和强化算法后提高到98.7%。在阻力训练中，测量的和估计的肌肉力量呈现相似趋势，屈伸的均方根误差（RMSE）为4.2 N，尺桡偏的RMSE为5.8 N。在两种不同速度和负荷下，动态肌肉力量的理论值和估计值呈现相似趋势，几乎没有相位差，与桡偏相比，屈伸运动期间的估计准确率更高。所提出的算法具有很强的通用性和实用性，旨在实现患者的自我康复训练。
