基于CMAC神经网络的能动磨盘智能控制实验研究

Experimental investigation on intelligent control of active lap based on CMAC neural network

针对能动磨盘面形控制系统的非线性和多变量特点,提出了基于CMAC神经网络的能动磨盘面形智能控制方法,以CMAC神经网络来映射磨盘面型和控制脉冲之间复杂的关系。为验证上述智能控制方法,搭建了由有效变形口径为420 mm能动磨盘和60路微位移阵列传感器组成的3单元能动磨盘面形检测实验平台,在该实验平台上进行了多组实验,利用微位移阵列传感器分别检测出能动磨盘在1单元、2单元和3单元驱动器作用下实验面形相对于理论面形的偏差,其中峰谷值分别为0.99,2.34和2.68μm,均方根值分别为0.19,0.59和0.57μm,实验结果验证了能动磨盘CMAC神经网络智能控制的可行性。

A new way to control the active lap by CMAC （cerebellar model articulation controller） neural network is put forward, based on the controlling system features of nonlinear and muti-variables. Using parameters of lap surface as input variables, pulse voltages of servo control unit as output variables, the function between input and output variables could be fitted by CMAC neural network. To testify this intelligent controlling idea, an experimental platform is constructed, which is composed of an active lap and lap testing system. The lap is controlled by 3 actuate motors with its available changing caliber 4420 mm, and the change of lap surface could be measured by 60 micro displacement sensor matrix. The active lap is actuated by a single motor, two motors and three motors respectively, and the errors between experimental and theoretical surface are measured. The peak-valley value are 0.99 μm, 2.34 μm and 2.68 μm, and the root mean square value are 0. 19 μm, 0.59 μm and 0.57 μm respectively. These results testify the feasibility of active lap controlled by CMAC neural networks.