AI-driven rehabilitation and assistive robotic system with intelligent PID controller based on RBF neural networks
AI-driven rehabilitation and assistive robotic system with intelligent PID controller based on RBF neural networks
复制标题
DOI：
10.1007/s00521-021-06785-y
复制
发表时间：
2022-01
影响因子：
6
通讯作者：
Wei Xiao;Kai Chen;Jiaming Fan;Yifan Hou;Weifei Kong;Guo Dan
中科院分区：
文献类型：
--
作者：
Wei Xiao;Kai Chen;Jiaming Fan;Yifan Hou;Weifei Kong;Guo Dan
本文设计了一种基于镜像治疗和虚拟刺激的合作双侧上肢康复机器人系统，用于辅助偏瘫患者进行康复训练。可以将偏瘫的患病肢体放置在带有伺服马达的机械臂上，而将健康肢体放置在没有伺服马达的另一侧。在机械臂的辅助下，患病肢体可以跟踪健康肢体进行镜像运动，完成康复训练。该机械臂装置可进行个人调节，辅助患者肘关节屈曲和腕关节旋转。以增强患者积极康复的意愿。开发了一种基于游戏的康复训练，实现人机交互，并提供视觉刺激。采用基于径向基函数（RBF）神经网络的自适应比例-积分-导数（PID）控制器来提高机械臂受影响侧的跟踪性能。RBF神经网络通过网络输出与系统输出之间的误差信号更新其参数。利用雅可比矩阵和健康侧与患病侧的运动误差对PID参数进行更新。通过实验验证，RBF-PID控制器在响应速度、抗干扰性、轨迹等方面都优于传统的PID控制器。分析了不同加载条件下的系统响应，并绘制了系统响应图。两侧对应关节角度的这些误差值可以解释为非常小。实践证明，该系统完成了康复训练，反映了患者的主动康复意识。
In this article, a cooperative bilateral upper-limb rehabilitation robotic system based on mirror therapy (MT) and virtual stimulation was developed to assist hemiplegia in rehabilitation training. The hemiplegia’s affected limb can be placed on one of the robotic arms with a servomotor, and the healthy limb is placed on another side without servomotor. With the assistance of the robotic arm, the affected limb can track the healthy limb to perform mirror motion to complete the rehabilitation training. The robotic arm device can be adjusted personally to assist the patient's elbow joint flexion and wrist joint rotation. In order to enhance the willingness of patients to actively recover. A game-based rehabilitation training was developed to realize human–computer interaction and to provide visual stimulation. The adaptive proportional–integral–derivative (PID) controller based on radial basis function (RBF) neural network has been adopted to improve the tracking performance of the affected side of robotic arm. The RBF neural network updates its parameters through the error signal between output of network and output of the system. The parameters of PID are updated by Jacobian matrix and the movement error between the healthy side and the affected side. Its abilities of RBF-PID controller about response speed, anti-interference and tracks are better than conventional PID controller’s through experimental validations. The system response was analyzed and graphed for different loading conditions. These error values of the angle of corresponding joint on both sides can be interpreted as very low. The system was proved to complete rehabilitation training and reflect the patient’s awareness of active rehabilitation.