基于机器视觉的PCB 光板缺陷检测技术研究

1、 （申请工学硕士学位论文） 基于机器视觉的PCB 光板缺陷 检测技术研究培养单位 ：信息工程学院 学科专业 ：通信与信息系统 研 究 生 ：胡文娟指导教师 ：刘 泉 教 授2007年4月基于机器视觉的PCB光板缺陷检测技术研究胡文娟武汉理工大学分类号密 级 UDC 学校代码学 位 论 文题 目 基于机器视觉的PCB 光板缺陷检测技术研究 英 文 Research of Machine Vision Based题 目 Defect Detection Techniques on PCB 研究生姓名姓 名职 称 单位名称 邮 编申请学位级别 硕士 学科专业名称 通信与信息系统 论文提交日期 200

2、7年4月 论文答辩日期 2007年5月 学位授予单位 武汉理工大学 学位授予日期 答辩委员会主席 陈 伟 评阅人 周 祖 德2007年4月指导教师摘 要印刷电路板(PCB是集成各种电子元器件的信息载体，在电子领域中有着广泛的应用。随着技术的不断发展和工艺水平的不断提高，电子产品趋于更轻、更薄、更小，PCB 朝着层数更多、密度更高的方向发展，这使得PCB 的质量检验成为一件非常困难的工作。传统的人工检测方法容易漏检、检测速度慢、检测时间长，已经不能满足生产的需要，如何更有效的实现PCB 的自动缺陷检测，成为半导体工业领域一个热门问题。机器视觉检测技术，集电子学、光电探测、图像处理和计算机技术于一

3、身，是精密测试技术领域内最具有发展潜力的新技术。机器视觉系统，一般采用CCD 或CMOS 摄像机摄取待检测目标并转化为数字信号，再采用先进的计算机硬件与软件技术对数字图像信号进行处理，从而得到所需要的各种目标图像特征值，并由此实现零件识别或缺陷检测等多种功能。然后再根据其结果显示图像，输出数据，由反馈信息引导执行机构完成位置调整、好坏筛选等自动化流程。将机器视觉引入到工业检测中，具有非接触、速度快、柔性好等突出优点，在现代制造业中有着重要的应用前景。本文将机器视觉技术应用到PCB 光板的缺陷检测中，实现PCB 光板的自动缺陷检测。在研究机器视觉技术的基础上，针对PCB 光板上的几种常见几何缺陷

4、类型，制定PCB 光板缺陷检测系统总体方案，讨论视觉检测系统工作原理，为PCB 光板视觉检测系统搭建硬件平台：包括照明系统、图像采集系统、以及控制台系统；重点针对采集的PCB 光板图像讨论视觉检测算法并进行仿真实验，包括图像预处理、分割、描述、数学形态学、模式识别等方法，着重根据PCB 设计规则运用数学形态学及模式识别方法完成自动检测识别；最后根据视觉检测算法设计系统软件，对PCB 光板完成缺陷的检测与识别。对包含不同缺陷的PCB 光板图像进行实验，结果证明本文的PCB 光板缺陷检测系统能够对PCB 光板上的短路、断路、毛刺、缺损四种主要缺陷做出有效的检测、定位与识别。关键字：印刷电路板(PC

5、B；机器视觉；数学形态学；模式识别AbstractPrinted Circuit Board, a kind of information carrier which integrates varieties of electronic devices, has popular applications in electronic fields nowadays. With the rapid development of manufacture techniques, electronic product tends to lighter, thinner and smaller, and

6、PCB turns to have more layers and higher density, which makes quality detection of PCB become more difficult. Traditional detection methods can not satisfy the large production for inaccurate, slow and long time detection, then how to realize automated defect detection of PCB becomes a hot topic in

7、semiconductor industry.Machine vision technology, which combines electronics, photoelectric detection, image processing and computer technology into oneself, is a potential new technology in industrial detection field. Machine vision system, usually obtains digital image signals of detected object b

8、y CCD or CMOS camera, then processes the digital image signals to get characteristic values by adopting advanced computer hardware and software techniques, and accomplishes workpiece recognition or defect detection accordingly. Based on the results, the system displays the images, exports the data a

9、nd sends out instructions to control corresponding equipment to act such as location adjusting and quality filtering according to feedback information.Machine vision technology is applied into automated PCB defect detection in this paper. On the basis of studying machine vision technology, we design

10、 the whole scheme of PCB defect detection system towards several simple geometric defects on PCB, discuss the principle of the system, and establish hardware platform for the system, including illuminating unit, image acquisition unit and control unit. Then, we discuss and design the foremost vision

11、 detection algorithm towards PCB image, including image pre-processing, segmentation, description, mathematical morphology, and pattern recognition, while the key is to accomplish defect detection using mathematical morphology and pattern recognition based on PCB design rules. Finally, we design sys

12、tem software according to the vision detection algorithm to realize defect detection and recognition on PCB.Experimental results demonstrate that though the PCB defect detection system described in this paper, four types of defects including short circuit, open circuit,protuberance and concavity on

13、PCB can be effectively detected, located and recognized.Keywords: Printed Circuit Board(PCB; Machine Vision; Mathematical Morphology;Pattern Recognition目 录第1章 绪 论·······················

14、··················································

15、··································11.1 机器视觉检测技术综述·············

16、83;·················································

17、83;················11.2 PCB光板视觉检测技术发展状况及分析·····························

18、3;·······················31.3 课题的来源、目的及意义························

19、··················································

20、··51.4 本文主要研究内容及组织结构·············································

21、·······················6第2章 PCB光板视觉检测系统总体设计·······················

22、3;········································72.1 问题的提出········

23、··················································

24、··········································72.2 检测系统工作原理······

25、;··················································

26、;································72.3 照明系统设计················

27、··················································

28、······························82.4 PCB图像采集系统设计·················&#

29、183;·················································&#

30、183;············92.4.1 CCD摄像机···································&

31、#183;·················································&

32、#183;··102.4.2 图像采集卡·············································

33、···········································122.5 控制台设计·····&

34、#183;·················································&

35、#183;··········································132.6 本章小结·····

36、83;·················································

37、83;·············································14第3章 视觉检测算法分析··&

38、#183;·················································&

39、#183;·································153.1 图像预处理··············&#

40、183;·················································&#

41、183;·································163.1.1 图像平滑··············&#

42、183;·················································&#

43、183;···························173.1.2 图像对比度增强····················

44、;··················································

45、;··········203.1.3 图像锐化······································

46、;··················································

47、;····243.1.4 采用的预处理方法···········································&

48、#183;································273.2 图像分割···············

49、83;·················································

50、83;···································273.2.1 最大类间方差法············

51、··················································

52、··················283.2.2 聚类阈值分割·····························

53、3;·················································

54、3;····303.2.3 迭代阈值分割···········································&#

55、183;········································313.2.4 采用的分割方法·······

56、;··················································

57、;·······················313.3 图像描述·························&

58、#183;·················································&

59、#183;·························333.3.1 邻接与连通······················

60、··················································

61、················333.3.2 线描述································&

62、#183;·················································&

63、#183;·············343.3.3 区域描述··································&

64、#183;·················································&

65、#183;·······353.3.4 模板匹配········································&

66、#183;·················································&

67、#183;·373.3.5 采用的描述方法·············································

68、3;··································383.4 二值形态学滤波·············

69、83;·················································

70、83;··························383.4.1 集合······················

71、;··················································

72、;···························393.4.2 二值腐蚀运算····················

73、83;·················································

74、83;·············393.4.3 二值膨胀运算··································&

75、#183;·················································4

76、03.4.4 二值开运算················································

77、········································413.4.5 二值闭运算········

78、;··················································

79、;······························423.4.6 基本性质··················

80、;··················································

81、;························433.4.7 连通区域标记·······················

82、83;·················································

83、83;··········433.4.8 采用的形态学滤波方法····································

84、83;·······························453.5 图像缺陷检测、定位与识别···············

85、83;·················································

86、83;····463.5.1 图像模式识别···········································&

87、#183;········································463.5.2 PCB图像识别······

88、3;·················································

89、3;···························473.5.3 采用的PCB 图像识别方法···················&

90、#183;···········································483.5.4 缺陷的检测····

91、··················································

92、··································483.5.5 短路、断路的识别·············&#

93、183;·················································&#

94、183;············533.5.6 毛刺、缺损的识别··································

95、3;·········································543.5.7 缺陷定位及识别总结······

96、;··················································

97、;················553.6 本章小结································&

98、#183;·················································&

99、#183;··················56第4章 系统软件设计·····························&

100、#183;·················································&

101、#183;··············574.1 开发工具选择·································&

102、#183;·················································&

103、#183;··········574.2 PCB视觉检测系统软件功能模块···································

104、3;···························574.2.1 系统主程序流程····················&

105、#183;·················································&

106、#183;·········584.2.2 缺陷检测流程······································

107、;··············································594.2.3 缺陷识别流程·

108、83;·················································

109、83;································604.3 PCB板缺陷视觉检测及识别实验示例·············

110、3;·········································624.3.1短路识别·······

111、;··················································

112、;····································624.3.2断路识别············

113、··················································

114、·······························634.3.3毛刺识别·················&

115、#183;·················································&

116、#183;·························644.3.4缺损识别······················&#

117、183;·················································&#

118、183;····················654.4 实验结果分析···························&#

119、183;·················································&#

120、183;················664.4.1 结果分析·······························&#

121、183;·················································&#

122、183;··········664.4.2 影响图像检测精度因素分析····································

123、························664.5 本章小结························&#

124、183;·················································&#

125、183;··························67第5章 总结与展望·····················

126、83;·················································

127、83;··························685.1 总结······················&

128、#183;·················································&

129、#183;····································685.2 展望············

130、;··················································

131、;···············································68 参考文献·

132、3;·················································

133、3;·················································

134、3;·············70 致 谢···································

135、3;·················································

136、3;·····························73 附录 攻读硕士期间发表的论文··················

137、;··················································

138、;··········74第1章 绪 论1.1 机器视觉检测技术综述检测技术是制造业的基础，随着制造水平的快速发展，制造领域不断扩大，产品质量不断提高。相应地，对检测技术提出了新的需要，传统意义上的很多检测方法已经不能适应现代制造业的要求。比如在汽车工业中，为全面控制车身的制造质量，需要在制造现场对制造过程中的产品(或零件 实行检测，这是一类典型的在线测量问题，在现代制造业中具有广泛代表性，也是传统测量方法难以解决的。对于工业中批量生产的产品，传统的人工检测存在以下几个不可避免的缺点：(1 容易

139、漏检。由于是人眼检测，眼睛容易疲劳，会造成故障不能被发现的问题。并且人工检测主观性大，判断标准不统一，使检测质量变得不稳定。(2 检测速度慢，检测时间长。比如对于图形复杂的印刷电路板，人工很难实现快速高效的检测，因此人工检测不能满足高速的生产效率。(3 随着技术的发展，设备的成本降低，人工费用增加，仍然由人工进行产品质量控制，将难于实现优质高效，而且还会增加生产成本。(4 在信息技术如此发达的今天人工检测有不可克服的劣势，例如：对检测结果实时地保存和远距离传输，对原始图像的保存和远距离传输等。(5 有些在线检测系统是接触式检测，需要与产品进行接触测量，因此，有可能会损伤产品。机器视觉(Mach

140、ine Vision检测技术，综合运用了电子学、光电探测、图像处理和计算机技术，是精密测试技术领域内最具有发展潜力的新技术。将机器视觉引入到工业检测中，实现对物体(产品或零件 三维尺寸或位置的快速测量，具有非接触、速度快、柔性好等突出优点，在现代制造业中有着重要的应用前景。机器视觉工业检测系统就其检测性质和应用范围而言，分为定量检测和定性检测两大类，每类又分为不同的子类。机器视觉在工业在线检测的各个应用领域十分活跃，如：印刷电路板的视觉检测、钢板表面的自动探伤、大型工件平行度和垂直度测量、容器容积或杂质检测、机械零件的自动识别分类和几何尺寸测量等。此外，在许多其它方法难以检测的场合，利用机器视

141、觉系统可以有效地实现。机器视觉的应用正越来越多地代替人去完成许多工作，这无疑在很大程度上提高了生产自动化水平和检测系统的智能水平1,2。具体来讲，机器视觉系统是指通过机器视觉产品，即图像摄取装置，将被摄取目标转换成图像信号，传送给专用的图像处理系统，根据像素分布和亮度、颜色等信息，转变成数字化信号；图像系统对这些信号进行各种运算来抽取目标的特征，进而根据判别的结果来控制现场的设备动作。机器视觉系统基本构成框图如图1-1所示。 图1-1 机器视觉系统基本构成框图由于机器视觉系统可以快速获取大量信息，而且易于自动处理，也易于同设计信息以及加工控制信息集成，因此，在现代自动化生产过程中，人们将机器视

142、觉系统广泛地用于工况监视、成品检验和质量控制等领域。机器视觉系统的特点是提高生产的柔性和自动化程度。在一些不适合于人工作业的危险工作环境或人工视觉难以满足要求的场合，常用机器视觉来替代人工视觉；同时在大批量工业生产过程中，用人工视觉检查产品质量效率低且精度不高，用机器视觉检测方法可以大大提高生产效率和生产的自动化程度。而且机器视觉易于实现信息集成，是实现计算机集成制造的基础技术。DVT 总裁Robert Steinke认为，“任何通过人来完成的检测过程，都适合于用机器视觉技术来代替”。总之，随着机器视觉技术自身的成熟和发展，可以预计它将在现代和未来制造企业中得到越来越广泛的应用。在国外，机器视

143、觉的应用普及主要体现在半导体及电子行业，其中大概40%-50%都集中在半导体行业。具体如PCB 印刷电路：各类生产印刷电路板组装技术、设备；单、双面、多层线路板，覆铜板及所需的材料及辅料；辅助设施以及耗材、油墨、药水药剂、配件；电子封装技术与设备；丝网印刷设备及丝网周边材料等。SMT 表面贴装：SMT 工艺与设备、焊接设备、测试仪器、返修设备及各种辅助工具及配件、SMT 材料、贴片剂、胶粘剂、焊剂、焊料及防氧化油、焊膏、清洗剂等；再流焊机、波峰焊机及自动化生产线设备。电子生产加工设备：电子元件制造设备、半导体及集成电路制造设备、元器件成型设备、电子工模具。机器视觉系统还在质量检测的各个方面已经

144、得到了广泛的应用，并且其产品在应用中占据着举足轻重的地位3-6。而在中国，以上行业本身就属于新兴的领域，再加之机器视觉产品技术的普及不够，导致以上各行业的应用几乎空白，即便是有，也只是低端方面的应用。目前在我国，随着配套基础建设的完善，技术、资金的积累，各行各业对采用图像和机器视觉技术的工业自动化、智能化需求开始广泛出现。国内有关大专院校、研究所和企业近两年在图像和机器视觉技术领域进行了积极思索和大胆的尝试，逐步开始了工业现场的应用。其主要应用于制药、印刷、矿泉水瓶盖检测等领域。这些应用大多集中在如药品检测分装、印刷色彩检测等，真正高端的应用还很少，因此，以上相关行业的应用空间还比较大。当然，

145、其他领域如指纹检测等等领域也有着很好的发展空间。1.2 PCB光板视觉检测技术发展状况及分析随着信息技术的发展，微电子产业已经成为信息产业的核心，而集成电路又是信息技术产业群的基础，因此发展集成电路产业是我国信息产业发展的重中之重。微电子产业是一个高新技术行业，同时也是一个巨大的行业，它包含了科研开发、生产加工、质量检测等诸多方面，而微电子产品的检测技术是产品研发及生产加工质量保证的关键技术。印刷电路板(PCB是集成各种电子元器件的信息载体，在各个领域得到了广泛的应用。随着技术的不断发展和工业的持续进步，电子产品趋于更轻、更薄、更短、更小，也使得PCB 制造技术朝更高密度发展。由于这些原因，生

146、产及更换它们的成本也越来越高。所以，需要相应的质量控制手段，使每一层上的线路都能够在上一层铺设之前被检查，排除或修复大部分缺陷。在PCB 光板的大批量生产过程中，出现的故障基本都是线路错误，主要可分为：短路、断路、毛刺、缺损四类。如果不及时地将这些质量问题检查出来，势必在PCB 板调试和使用过程中留下隐患，造成更大的损失7,8。自动光学检测(AOI, Automated Optical Inspection是近几年兴起一种视觉检测方法。它是通过CCD 照相的方式获得器件的图像，然后经过计算机的处理和分析比较来判断缺陷和故障。其优点是检测速度快，编程时间较短，可以放到生产线中的不同位置，便于及时

147、发现故障和缺陷，使生产、检测合二为一，可缩短发现故障和缺陷时间，及时找出故障和缺陷的成因。因此它是目前采用得比较多的一种检测手段。在国外，经过10多年的努力，自动光学检测系统(AOI最终被成功地运用在印刷电路板生产线上。在这段时间内，AOI 供应商的数量急剧增加，各种AOI 技术也得到了长足发展。目前，从简单的摄像系统到复杂的3-D X光检测系统，众多供应商们已经几乎能够提供可以适用于所有自动生产线的AOI 设备9。目前具备AOI 系统的供应商有英国Diagnosys 公司的Vision Point型，由中国三吉电气(集团 有限公司代理；美国Tera Dyne公司的5500型，由香港DEI 公

148、司代理；美国Angilent 公司的SDX(有X 射线 ，由安捷伦公司代理。此外还有其它PCB 自动检测仪，如Sony Minokamo Corporation(索尼美浓加茂株式会社 的CPC-1000系列产品，Orbotech 公司生产的Trion-2000系列。对于Trion-2000系列，由于采用了多摄像头技术，因此对于很多平常难以检测到的缺陷都可以发现，功能十分强大，但价格十分昂贵，而国内还没有这样强大功能的自主产品。随着科学技术的飞速发展和工业自动化程度的提高，高速、高精度、非接触的在线检测已成为检测行业的发展方向，它可以大大地解放劳动力，达到提高生产效率和产品质量、降低成本的目的。

149、科技的发展的和使用要求的提高，对应用于办公自动化和产品质量检测的图像输入、检测和识别系统提出了大幅面、高速、高分辨率和高精度的要求。在办公自动化系统中，要求对大幅面图纸等进行高速、高精度的扫描输入，以提高办公自动化的效率和质量；在产品质量检测中，要求对宽幅打印机打印质量(高分辨率 、彩色印刷套色质量(印刷品的表现力 、PCB 制造质量和液晶点阵显示器质量检测(加工制造密度己经达到近10微米的数量级 、硬盘磁头等进行高精度检测和跟踪控制。假如以前依靠人眼进行质量检测，现在若仍采用该手段，已无法达到质量检测目的10-12。机器视觉技术与图像扫描技术密不可分，扫描技术已广泛应用于将光学信号转变为电信号的设备中。光学技术的发展，使得人类可以借助于摄像机、照相机、显微镜等获取的图像信息记录、观察和描述客观世界。CCD 技术是一种基于数字光学的扫描技术，其品质的高低在很大程度上就决定了扫描图像的质量。CCD 器件的出现使扫描技术由原来的机械扫描转向电子扫描，是扫描技术的一大发展。利用CCD 器件扫描不仅可以大大提高扫描速度，而且使扫描更稳定可靠、易于控制且自动化操作。利用数字扫描机或数字照相机对景物图像、图形、文件等进行扫描拍摄，将光影像聚焦到扫描机内光电荷藕合器件CCD 传感器上进行光电转换成数字电信号，再把电信号转输到半导体存