为了提高扫描隧道显微镜微位移工作台的定位精度,提出了一种基于遗传算法的神经网络PID控制方案.微位移工作台以压电陶瓷为驱动器、柔性铰链为导向机构,在分析工作原理的基础上,建立了工作台的数学模型.神经网络PID控制器对工作台进行闭环控制,能够在线调整网络加权值,实时改变PID控制器的系数,减小工作台的位移误差.利用遗传算法的全局搜索能力对BP网络的初始权值进行学习优化,有效消除了神经网络对初始权值敏感和容易局部收敛的缺陷,改善了控制器的控制效果.性能测试表明,12μm阶跃参考输入下的稳态误差从3.24%减小到2.55%,稳态时间从1.7 s缩短到1.1 s.
A PID control scheme based on the genetic algorithm and neural network was proposed to improve the position accuracy of the micro-displacement stage of scanning tunneling microscope. The stage was actuated by the piezoelectric ceramic actuators and the flexible hinges were used as its guiding mechanism. After the principle of the stage was analyzed, its mathematical model was established. Because the stage was controlled by the neural-network PID controller in a closed routine, the weights of BP network were adjusted on-line and the parameters of PID controller can be changed in real-time to reduce the displacement error of stage. The initial weights were optimized by the global property of genetic algorithm to overcome the initial weight sensitivity and local convergence of neural network and improve the control effect of the PID controller. The tested results show that the stable error of a step of 12 μm is reduced from 3.24% to 2.55% and the response time is shortened from 1.7 s to 1.1 s.