CASIA OpenIR  > 毕业生  > 硕士学位论文
基于机器视觉的钕铁硼表面缺陷检测系统
其他题名NdFeB Surface Defects Detection System Based on Machine Vision
吴亮
学位类型工学硕士
导师王欣刚
2012-05-21
学位授予单位中国科学院研究生院
学位授予地点中国科学院自动化研究所
学位专业模式识别与智能系统
关键词目标分割 特征提取 分类识别 凸缺陷分析 图像细化 Object Segmentation Feature Extraction Classification And Recognition Convexity Defects Analysis Image Thinning
摘要钕铁硼磁性材料是迄今为止性价比最佳的磁性材料。利用钕铁硼磁性材料生产的钕铁硼磁体,是当前稀土永磁材料行业的主导元件产品。其不仅应用于家用电器,而且广泛应用于航空、航天、电子、汽车工业、石油化工、仪器仪表及其它需要永久磁场的装置和设备中。钕铁硼产业作为朝阳产业,稀土永磁钕铁硼材料的新应用、生长点在不断涌现,钕铁硼材料产业市场前景十分广阔。然而,由于目前钕铁硼磁体加工的工艺、设备水平有限,加之电镀造成的磁体表面损伤,钕铁硼磁体的成品率很低。使用存在表面缺陷的磁体,不仅会严重影响整机设备的性能、寿命,而且会造成难以预估的经济损失。目前,该领域的表面检测以人工目测为主,无法达到产品质量和生产效率两方面的要求。 本文基于对钕铁硼表面缺陷类型的分析,提出了一种有效的检测算法,设计了一套基于机器视觉的钕铁硼表面缺陷检测系统。具体地,本文完成的工作和取得的研究成果主要包括以下4个方面: (1)按照模块化设计思想,我们将系统分为自动上料模块,工件传输、翻面和旋转模块,图像采集模块,输入输出控制模块,人机界面模块和图像处理模块。其中,输入输出控制模块中,设计了相机软触发控制和正品次品分选控制流程;图像处理模块分为缺陷目标分割、缺陷特征提取和正品次品分类识别三个子模块。 (2)缺陷目标分割:比较了三种自适应阈值分割算法,针对本文处理图像,提出了改进的OTSU分割算法。首先,采用投影法获得缺陷目标的区域,统计该区域的直方图h(x),像素个数为n;然后,在直方图中背景均值灰度等级u处,增加同等比例像素统计,即h(u)=h(u)+n;最后,对修正后的直方图采用OTSU算法分割图像,其分割效果比一般的OTSU算法好。 (3)缺陷特征提取:首先,针对异形缺陷,给出了Hough变换直线检测和多边形逼近算法,来计算工件的长宽比和对边平行度参数,并作为异形缺陷的评价参数;其次,针对缺角缺陷,进行了工件凸包的凸缺陷分析和直线拟合误差分析,采用工件凸包的凸缺陷的最大深度和次大深度,以及直线度拟合误差对缺角缺陷进行评价,其中,采用整体最小二乘法进行的直线拟合误差可以评价碎边缺陷的严重程度;再次,针对麻坑缺陷,提出了基于工件区域直方图统计和麻坑面积大小先验知识的阈值分割方法,提取最大连通域面积作为麻坑的评价参数;最后,针对裂纹缺陷,采用数学形态学中的膨胀、腐蚀、细化等方法提取了裂纹特征。 (4)正品次品分类识别:基于提取的缺陷特征,设计了BP神经网络和简单分类器相结合的串联组合分类器,进行正品次品分选。
其他摘要NdFeB has the highest performance price ratio of all magnetic materials. The NdFeB magnets, made of NdFeB magnetic materials, are the leading elements of rare-earth permanent magnetic materials industry. It’s not only used in household appliances, and is widely used in aviation, aerospace, electronics, automobile industry, petrochemical industry, instrumentation and other installations and equipment in need of permanent magnetic field. NdFeB industry is a sunrise industry, the new application and growing point of which are coming forth. NdFeB materials industry’s market prospect is very broad. However, because of limited equipment and technology of NdFeB magnet processing at present and surface damage caused by electroplating, the yield of NdFeB is very low. Using magnets with surface defects, will not only seriously affect the performance and life of the whole machine, and lead to difficult to estimate economic losses. Now, the surface defects detection in the field is mainly by manual, which is unable to meet the requirements of both product quality and production efficiency. In this paper, based on analysis of the NdFeB surface defects type, we proposed an efficient detection algorithm and designed a set of NdFeB surface defects detection system based on machine vision. Specifically, in the paper, the work that has been done and research achievements are as follows: (1) According to the modular design, our system is divided into the following modules: automatic feeding module, workpiece’s transmission, turn-over and rotation module, image acquisition module, I/O control module, in which, we designed the process of camera soft trigger control and sorting control, man-machine interface module and image processing module, which includes defective object segmentation, defective feature extraction and classification and so on. (2) Defective object segmentation: we compared 3 kinds of adaptive threshold segmentation algorithm, and in view of our image, proposed an improved OTSU segmentation algorithm. First, adopt projection method to get the defect object region (the number of pixels is n) and compute the histogram h (x) of this region; second, add equal proportion pixel number at u, which is the mean gray level of background, in the histogram, that is, h (u) =h (u) + n; finally, based on the new histogram, segment the image with OTSU method, and the result is better than general OTSU algorithm. (3) Defective feature extraction: first, in view of abno...
馆藏号XWLW1783
其他标识符200928014628061
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/7620
专题毕业生_硕士学位论文
推荐引用方式
GB/T 7714
吴亮. 基于机器视觉的钕铁硼表面缺陷检测系统[D]. 中国科学院自动化研究所. 中国科学院研究生院,2012.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
CASIA_20092801462806(3203KB) 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[吴亮]的文章
百度学术
百度学术中相似的文章
[吴亮]的文章
必应学术
必应学术中相似的文章
[吴亮]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。