用神经网络结构实现超磁致伸缩智能构件滑模控制  

Implementation of sliding mode control of giant magnetostrictive smart component by neural network

作　　者：赵章荣[1,2] 邬义杰[1] 顾新建[1] 张雷[1] 王彬[1] 

机构地区：[1]浙江大学现代制造工程研究所流体传动与控制国家重点实验室,浙江杭州310027 [2]华北科技学院机电系,河北燕郊101601

出　　处：《光学精密工程》2009年第4期778-786,共9页Optics and Precision Engineering

基　　金：国家自然科学基金资助项目(No.50575205);国家"863"高技术研究发展计划资助项目(No.2006AA04Z233,2007AA04Z101);国家教育部博士点基金资助项目(No.20070335204);浙江省自然科学基金重点资助项目(No.Z1080537)

摘　　要：提出了一种利用超磁致伸缩材料(GMM)智能构件精密加工活塞异形孔的方法。采用一种神经网络前馈复合离散滑模变结构控制策略,实现GMM智能构件的精密位移控制,消除了GMM智能构件迟滞非线性的影响。将智能构件的输出位移及其变化率作为小脑模型神经网络(CMAC)输入,构件的输入电流作为网络输出,利用CMAC在线自学习能力建立GMM智能构件的迟滞逆模型;通过离散滑模变结构控制器来消除神经网络的建模近似误差以及外界干扰。仿真结果表明,此控制策略能在线建立智能构件的迟滞逆模型,消除迟滞非线性的影响,控制误差降低到1.5%以内,可实现智能构件的精密位移控制。A new method for precise machining non-cylinder pin holes of pistons was presented by using Giant Magnetostrictive Material（GMM） smart components. To eliminate the impact of GMM smart component hysteresis and nonlinearity, a real-time hysteretic compensation control strategy combining a CMAC neural network feedforward controller and a sliding mode controller was proposed to implement the precision position tracking control of the smart component. The output and the output rate were used as the input data of CMAC neural network of the current smart component, the input current as the output of the neural network,and an inverse hysteresis model based on the GMM smart component was established by the CMAC network on-line learning. The model approximate error of CMAC neural network and the external disturbance were eliminated by using the discrete sliding controller. Simulation results show that the control strategy could on-line obtain the inverse hysteresis model of the smart component with a controll error less than 1.5 %, and could eliminate the hysteretic nonlinear impact on the smart component.

关 键 词：超磁致伸缩智能构件 小脑模型神经网络(CMAC) 滑模变结构控制 前馈补偿 迟滞非线性 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP271.7[自动化与计算机技术—控制科学与工程]