人工肌肉系统神经网络建模与控制  

MODELING AND CONTROLLING OF ARTIFICIAL MUSCLE SYSTEM BASED ON NEURAL NETWORKS

作　　者：田社平[1] 丁国清[1] 颜德田[1] 林良明[1] 

机构地区：[1]上海交通大学信息检测技术及仪器系,上海200030

出　　处：《中国生物医学工程学报》2003年第4期300-308,共9页Chinese Journal of Biomedical Engineering

基　　金：中国科学院机器人学开放实验室资助项目

摘　　要：针对广泛应用于医疗机器人等领域的空气压人工肌肉的非线性特性 ,运用神经网络模型 ,对人工肌肉系统进行建模与控制。设计了一个具有单关节的人工肌肉试验系统。采用递推预报误差 (RPE)算法来训练神经网络 ,并给出了RPE算法的具体实现 ,该算法与传统的BP算法相比具有更好的收敛特性。讨论了不同的输入向量和不同的隐层节点数对建模结果的影响。在此基础上提出了一种基于神经网络的人工肌肉系统非线性控制结构 ,并给出了人工肌肉关节跟踪正弦波和矩形波参考信号的试验结果。试验结果表明 ,采用神经网络模型对人工肌肉系统进行建模与控制 ,能有效地克服人工肌肉系统的非线性 ,能够得到更快的控制速度和更高的控制精度。The pneumatic artificial muscles are widely used in the fields of medical robots etc. Neural networks are applied to modeling and controlling of artificial muscle system. A single-joint artificial muscle test system was designed. The recursive prediction error (RPE) algorithm which is much more effective than back propagation (BP) algorithm was applied to train the neural networks. The realization of RPE algorithm was given. The difference of modeling of artificial muscles using neural networks with different input nodes and different hidden layer nodes was discussed. On this basis the non-linear control scheme using neural networks for artificial muscle system has been introduced. The experimental results show that the nonlinear control scheme is more effective than the traditional linear control scheme.

关 键 词：人工肌肉 神经网络 递推预报误差算法 非线性建模与控制 

分 类 号：R318[医药卫生—生物医学工程] TP271[医药卫生—基础医学]