摘要
分析了坐标测量机几何误差的几种常用模型,提出了基于神经网络的单项几何误差模型。由于坐标测量机几何误差变化规律复杂,采用一般的BP神经网络模型算法,速度慢且难以收敛。利用牛顿变形算法训练网络,加快了网络收敛速度,效果显著。通过与线性插值、多项式拟合法和神经网络逼近法的比较,可以明显看出用该神经网络算法的优越性。
Several coordinate measuring machine geometry error models of several kinds in common use are analyzed in this paper. A single geometry error model based on NN is presented. Owing to the complicated variable rule of CMMs geometry error,it's difficult to convergence for using common BP neural network model arithmetic with a slow velocity. This paper uses Newton transfiguration arithmetic to train NN so that the convergence of NN is aclelerated with remarkable effects. Compared with linear inserting value and multinomial imitation method,it is obvious that NN has more advantages.
出处
《西安理工大学学报》
CAS
2004年第2期164-167,共4页
Journal of Xi'an University of Technology
