摘要
为提高压电陶瓷驱动的微定位工作台的模型精度,提出了一种基于动态递归神经网络的建模方法。压电陶瓷具有极高的位移分辨率,但存在着迟滞非线性。分析了压电陶瓷驱动器的结构和特性,利用动态神经网络的自反馈结构和自学习能力,建立起工作台的网络模型,通过在线调整模型结构和参数,减小了工作台的建模误差。测量工作台的定位数据对网络模型进行了训练,实验结果表明,当工作台最大行程为80μm时,平均定位误差0.07μm,最大误差0.09μm,比采用静态网络模型有了一定的提高。
In order to improve the model accuracy of micro-positioning stage driven by piezoelectric ceramics, a new modeling method based on the dynamic recurrent neural network was proposed. The piezoelectric ceramics had super high resolution, but it had the property of hysterics and nonlinearity. The structure and characteristic of the stage were analyzed. The model was established based on the self-feedback and self-learning of the DRNN. The errorof the model was reduced by adjusting the structure and parameters online. Stage positioning data were used to train the net. The results of experiment showed that the average error and the maximum error within the journey of 80 μm were reduced to 0.07μm and 0.09 μm respectively. The poisoning precision was improved compared with the static neural network model.
出处
《压电与声光》
CSCD
北大核心
2010年第2期247-250,共4页
Piezoelectrics & Acoustooptics
基金
国家自然科学基金重大研究计划基金资助项目(No.90307003)
山东省教育厅科技计划基金资助项目(J08LJ89)
关键词
纳米定位
精密工作台
压电驱动器
系统辨识
神经网络
nano positioning
precision stage
piezoelectric actuator
system identification
neural network