基于BP神经网络的车削表面粗糙度检测  

Measurement of the Turning Surface Roughness Based on BP Neural Network

作　　者：闵莉[1] 王哲[1] 高龙飞[1] 刘大任[2] 

机构地区：[1]沈阳建筑大学机械工程学院,辽宁沈阳110168 [2]沈阳建筑大学学报编辑部,辽宁沈阳110168

出　　处：《沈阳建筑大学学报（自然科学版）》2015年第5期906-915,共10页Journal of Shenyang Jianzhu University：Natural Science

基　　金：国家自然科学基金项目(51375317)

摘　　要：目的研究基于计算机视觉的非接触式车削表面粗糙度检测方法.方法通过由数字显微镜、高分辨率摄像机和计算机等构成的图像采集系统获取车削工件表面图像,提取基于灰度共生矩阵(GLCM)的14个表面纹理特征参数,探讨各纹理特征参数与粗糙度Ra之间的变化规律;以14个表面纹理特征参数为输入层节点输入量,以粗糙度Ra为输出量,构建车削表面粗糙度BP神经网络检测模型.结果检测结果与实测值的绝对误差平均值不超过0.08μm,且相对误差平均值小于2%,BP神经网络检测模型具有较好的检测效果,取得了较高的检测精度.结论基于BP神经网络的车削表面粗糙度检测模型,能够满足表面粗糙度测量的精度要求,对于快速非接触式检测车削表面粗糙度的研究具有借鉴意义.This paper studied a measurement method of the non-contact turning surface roughness based on computer vision. The surface image of turning workpiece was obtained by the image collection system,which consists of digital microscopy,high-resolution camera and computer. 14 texture feature parameters were extracted based on the gray level co-occurrence matrix（ GLCM）. Taking these feature parameters as input data and the surface roughness Raas output,a BP neural network for the turning detection model was built. The result showthat the absolute average error is not more than 0. 08μm between test results and actual value,and the average of relative error is less than 2%. It is found that a good measuring result of the turning surface roughness can be obtained by the proposed model. Conclusion is that the measurement model based on BP neural network for turning surface roughness meets precision requirements. It has a great significance for study of the fast non-contact workpiece turning surface roughness.

关 键 词：纹理分析 灰度共生矩阵 表面粗糙度Ra BP神经网络 

分 类 号：TG84[金属学及工艺—公差测量技术] TP183[自动化与计算机技术—控制理论与控制工程]