柔性关节摩擦和不确定补偿的小波神经——鲁棒复合控制  

Hybrid Control Using Wavelet Neural Networks and Robust Method for Flexible Joint with Friction and Uncertainties Compensation

作　　者：孙汉旭[1] 褚明[1] 贾庆轩[1] 

机构地区：[1]北京邮电大学自动化学院,北京100876

出　　处：《机械工程学报》2010年第13期68-75,共8页Journal of Mechanical Engineering

基　　金：教育部科学技术研究重大资助项目(307005)

摘　　要：研究柔性关节的高精度位置跟踪问题。建立柔性关节二阶级联动力方程,并引入非连续摩擦力和干扰项,针对系统存在的柔性与摩擦并存、不确定参数摄动、外部干扰等复杂动力特性,提出一种复合控制策略。全局采用李雅普诺夫函数的Backstepping方法设计具有柔性补偿和L2干扰抑制性能的控制器;同时,针对常规神经网络无法辨识非连续性函数的问题,提出在局部采用小波神经网络对摩擦和不确定项进行补偿。该方法不仅避免了复杂的求导运算,而且无须检测关节角加速度和预知摩擦及不确定上界。闭环稳定性分析和数值试验表明,设计的控制器具有鲁棒稳定性和干扰抑制性能,可保证系统的跟踪误差和网络权值误差一致终值有界,且避免了柔性关节因摩擦引起的爬行和平顶现象。The high-precision position tracking control for flexible joint is researched.The second-order cascade dynamics equations of flexible joint are established;then,non-consecutive friction and external interference are introduced.In view of the complex dynamic character istics that flexibility coexists with friction,uncertain parameter perturbations and external disturbances,a hybrid control strategy is proposed.The global controller is designed by using Lyapunov function backstepping method to compensate the flexibility and suppress disturbance based on L2 property.In view of the problem that it is hard for the conventional neural networks to identify the non-consecutive function like friction,the wavelet neural networks approach is locally utilized to estimate the nonlinear terms including friction and uncertainties.This strategy can avoid the complicated derivation operation,joint angle acceleration measurement and upper bound forecast.The closed-loop stability analysis and numerical simulation show that the proposed controller has robust stability and disturbance attenuation performance.The tracking error and network weights error are uniformly ultimately bounded,and the crawling and flat top phenomena induced by friction are avoided.

关 键 词：柔性关节 非连续摩擦 BACKSTEPPING方法 小波神经网络 复合控制 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]