摘要
为解决齿廓图像边缘光学失真影响齿轮视觉测量精度问题,采用双-K-交叉验证方法改进支持向量回归算法(SVR)的参数寻优,以提高SVR拟合精度、实现基于局部支持向量回归的齿廓边缘失真补偿算法(L-SVR)。根据渐开线直齿圆柱齿轮相邻同名齿廓相似性,在齿廓局部失真区域选取齿廓失真区域待补偿信号的相邻同名齿廓极径rk、齿廓边缘信号μk、齿廓平均迹线εk、齿廓边缘信号动态分量υk组成L-SVR算法输入自变量,分别获取齿廓边缘失真区域补偿信号υk*、εk^*、μk^*。实验表明,补偿信号与该齿廓经过严格清洗后未出现边缘失真区域时的测量信号具有极强的相关性。基于L-SVR补偿算法的测量结果接近齿轮测量中心测量结果、相对误差更小,能够满足齿轮视觉测量的精度要求。
In order to solve the problem that the edge optical distortion of the tooth profile image affects the gear vision measurement accuracy,a tooth profile edge distortion compensation algorithm based on local support vector regression(L-SVR)is proposed.According to the similarity of the adjacent tooth profile of the involute spur gear with the same name,the algorithm selects the pole diameter rk,the tooth profile edge signalμk,the tooth profile average traceεk,and the tooth profile edge signal dynamic componentυk to form the input independent variable of the L-SVR algorithm.The L-SVR algorithm parameter optimization is realized through the double-K-cross verification method,and the compensation signalsυk^*,εk^*,μk^*of the edge distortion area of the tooth profile were obtained respectively.Experiments show that the compensation signal has strong correlation with the measurement signal when the tooth profile is strictly cleaned and there is no edge distortion area.The detection result based on the local SVR compensation algorithm is close to the measurement result of the gear measurement center,and the relative error is smaller,which can satisfy the precision requirements of gear vision measurement.
作者
孙禾
赵文珍
赵文辉
段振云
SUN He;ZHAO Wen-zhen;ZHAO Wen-hui;DUAN Zhen-yun(College of Mechanical Engineering,Shenyang University of Technology,Shenyang 110870,China;Department of Electrical and Information Engineering,Liaoning Institute of Science and Technology,Benxi 117004,China)
出处
《控制工程》
CSCD
北大核心
2020年第11期2030-2037,共8页
Control Engineering of China
基金
十二五国家科技支撑计划(2014BAF08B01)
辽宁省自然科学基金指导计划(20170540474)
特殊环境机器人技术四川省重点实验室开放基金(19kftk03)。
关键词
边缘失真补偿
局部支持向量回归
相似性
齿廓偏差
Edge distortion compensation
local support vector regression
similarity
tooth profile deviation