上海交通大学学报(自然版) ›› 2012, Vol. 46 ›› Issue (06): 931-935.

• 自动化技术、计算机技术 • 上一篇    下一篇

基于分组数据处理神经网络气动人工肌肉迟滞特性

 崔霞1, 施光林2, 沈伟3   

  1. (1.上海师范大学 天华学院,上海 201815; 2.上海交通大学 机械与动力工程学院,上海 200240; 3.上海海洋大学 工程学院,上海 201306)
  • 收稿日期:2012-02-17 出版日期:2012-06-28 发布日期:2012-06-28
  • 基金资助:

    上海市高校培养优秀青年教师科研专项基金(thc1005)

Study on Hysteresis of Pneumatic Artificial Muscle Based on Group Method of Data Handling Neural Network

 CUI  Xia-1, SHI  Guang-Lin-2, SHEN  Wei-3   

  1. (1. Tian Hua College, Shanghai Normal University, Shanghai 201815, China;2. School of Mechanical Engineering, Shanghai Jiaotong University, Shanghai 200240, China;3. College of Engineering, Shanghai Ocean University, Shanghai 201306, China)
  • Received:2012-02-17 Online:2012-06-28 Published:2012-06-28

摘要: 气动人工肌肉的动态特性中存在着非常复杂的迟滞现象.目前对其迟滞特性的研究很不充分,甚至对其输入空间都难以确定.为此,建立了单自由度气动人工肌肉实验平台,利用分组数据处理神经网络独特的自组织特性,运用数据挖掘技术探索气动人工肌肉迟滞特性的输入空间.将自适应模糊小脑模型神经网络引入滑模控制,基于已确定的输入空间,在每个采样周期逼近迟滞力不断变化的动态值,在线实时补偿迟滞力的影响.实验结果验证了输入空间选取的合理性和有效性.

关键词: 分组数据处理神经网络, 气动人工肌肉, 迟滞力, 输入空间

Abstract: There exits hysteresis phenomenon in the dynamics of pneumatic artificial muscle(PAM). By now the study on the hysteresis of PAM is so insufficient that even its input space is unknown. So, a single degree of freedom PAM experiment facility was established. And the input space for the hysteresis of PAM was explored via a data mining technique—group method of data handling(GMDH) neural network with the character of self organization.  Based on the input space, an adaptive fuzzy cerebellar model articulation controller(CMAC) neural network was introduced into sliding mode control and the hysteresis of PAM was online compensated in real time. The experimental results suggest the rationality and effectiveness of the input space for the hysteresis of PAM.

Key words: group method of data handling (GMDH) neural network, pneumatic artificial muscle cerebellar model articulation controller (PAM), hysteresis, input space

中图分类号: