基于神经网络的麻花钻圆度误差的检测  

Test of Twist Drill's Circularity Errors Based on Neural Network

作　　者：葛动元[1,2] 姚锡凡[1] 蒋寿生[2] 

机构地区：[1]华南理工大学机械与汽车工程学院,广东广州510640 [2]邵阳学院机械与能源工程系,湖南邵阳422004

出　　处：《中北大学学报（自然科学版）》2011年第3期286-291,共6页Journal of North University of China(Natural Science Edition)

基　　金：国家863高技术研究发展计划资助项目(2007AA04Z111);湖南省自然科学基金资助项目(09JJ6092)

摘　　要：麻花钻圆度误差检测时,首先采用万能工具显微镜测出其棱边投影采样点的坐标,然后将神经网络算法引入到相关的数据处理中来,以拟合出其棱边投影的椭圆表达式系数.在神经网络训练时,以钻头棱边投影的坐标及常数1作为网络的3路输入,性能指标为该网络输出与常数1的差值的平方;根据负梯度下降法来调整隐层神经元与输出神经元之间以及输入神经元与隐层神经元之间的连接权值,达到预定的性能指标时,获得一组稳定的权值,从中求得钻头棱边投影椭圆表达式的中心坐标与长短轴平方的倒数,据此即可求得麻花钻棱边投影的椭圆表达式系数;最后根据常规的方法求出其较准确的圆度误差.实验中提出的采用神经网络的权值拟合椭圆的表达式系数的算法,是工件形位误差数据处理的一种新方法.To evaluate the circularity error of a twist drill, the coordinates of sampling points of the margin projection were measured with universal tool microscope. The neural network was introduced into data processing in order to fit the elliptical coefficients of the margin projection. The neural network was trained with the coordinates of sampling points and constant 1 which were taken as 3 inputs, and the square of errors between the neural output and constant 1 as performance index. The weights between hidden neurons and output neuron, and weights between inputs neuron and hidden neurons were tuned, thus stable weight values could be obtained in the light of negative gradient descent mean until the desired performance index was reached, from which the central coordinates and reciprocals of elliptical long axis^s and short axis＇s square were obtained respectively, thus the whole expression of the margin projection ellipse were gotten. The circularity errors of twist-drills could be got in higher precision by adopting conventional means. The proposed algorithm adopting neural network weight to fit elliptical expression coefficients is a news dada processing means for measurement geometrical errorsof workpiece.

关 键 词：神经网络 麻花钻 圆度误差 性能指标 李雅普诺夫函数 

分 类 号：TP202.2[自动化与计算机技术—检测技术与自动化装置]