【作者】 吴彰良；

【导师】 孙长库；

【作者基本信息】 天津大学 ， 测试计量技术及仪器， 2013， 博士

【摘要】 随着工业生产自动化程度的不断提高，如何实现对多品种、大批量生产过程的实时在线检测一直是工业领域的难题和企业生产追求的目标。本课题结合油封企业的生产现状，依据企业产品质量标准，针对其提出的油封产品在线检测具体指标要求，设计搭建了油封缺陷的在线图像检测系统，以图像拼接与ROI分割、缺陷边缘检测以及缺陷分类识别为主线，对油封表面缺陷的图像检测技术理论与方法进行了系统深入的研究，提出了一系列针对性的、快速可靠的图像处理算法，形成了较为完整的油封缺陷在线检测的方法体系，并通过实验进行了对比分析。本文的主要研究工作如下：1）结合油封外观质量标准及企业生产要求，制定了油封在线图像检测系统的性能量化指标；设计了基于工业面阵CCD相机的远心光路图像采集系统、LED环形低角度照明系统以及油封夹持旋转机构，实现油封幅面圆周等分割成像采集，以满足高速、宽尺寸范围、高分辨率的检测要求，解决了成像视场范围和分辨力与系统硬件要求之间的优化配置问题。2）研究了图像配准和融合不同技术方法，针对油封环带等分子图像采集过程特点，借助旋转机构精密传动子系统的先验知识，通过刚性变换实现大尺寸油封的序列子图像的完整无缝拼接；考虑到油封不同环带区域的质量要求不一样，同一类型缺陷的量化指标差异较大，结合圆参数拟合和先验知识实现对油封图像ROI分割，以便后续图像处理。3）针对油封表面缺陷的特点及系统指标要求，研究了油封缺陷边缘检测的不同方法，并做了对比分析。利用小波变换的良好时频分析特性，提出了改进阈值的小波局部模极大值油封边缘检测方法；同时由于彩色图像的包含更多的边缘信息，提出了基于不同彩色空间的油封缺陷边缘检测方法；不同方法有着不同的特性表现。4）在分析油封表面不同类型缺陷的特点的基础上，定义了各类缺陷的不同特征参数空间，利用主分量分析方法对油封缺陷进行了有效特征抽取；研究探讨了支持向量机分类识别方法，对油封端面及唇口区域分别构建分类器，实现油封表面缺陷分类识别。更多还原

【Abstract】 With the continuous improvement of the degree of industrial productionautomation, how to achieve the real-time online testing in the process of multi-species,mass production has long been the problem in industrial fields and the goal pursuedby production enterprises. Reference to product quality standards and combining withthe actual production situation of enterprises, this project is developed to detectsurface defects for oil seal based on image inspection technique. In view of thespecific requirements of oil seal online production and testing, a computer visioninspection system is set up, the theory and method of image detection technology foroil seal surface defects is discussed detailly, including ROI segmentation, imagemosaic, defect edge detection, defects classification and identification, etc. The mainwork included in the dissertation is shown as follows:1) According to oil seal surface quality standards and production requirements,the performance of oil seal defects testing system is quantified concretely. An onlineimage inspection system is designed and built, mainly including the image acquisitionsystem baed on telecentric lens and industrial CCD camera, the low angle ring LEDlighting system, and the oil seal clamp rotation mechanism. This scheme can achieveoil seal circumference equal portions image acquisition, fit the testing requirements ofhigh-speed, wide size range and high resolution, and optimize configuration of fieldof view, resolution and hardware.2) With a view to the characteristics of oil seal image decile acquisition process,a simple method is taken to realize oil seal sub-images mosaic with priori knowledgeand rigid transformation. Taking into account the difference of the qualityrequirements of oil seal different ring regions, that is the same type of defect indifferent region being quantified differently, oil seal ROI segmentation can solve theproblem and simply the following image processing.3) Considering the particularity of oil seal surface defects and systemspecifications, a few different approaches to detect oil seal defects’ edge are putforward in detail and their effects comparison are analyzed. An improved thresholdwavelet local modulus maxima edge detection method is presented owing to thetime-frequency performance of wavelet transform. Meanwhile because of color imagecontaining more edge information, the oil seal surface defects detection methods based on different color space are advanced. Different oil seal defects’ edge detectionapproaches have their own particularity.4) In connect with the speciality of various surface defects of oil seal, thedissertation not only defines the feature parameter space of defects’ type, but alsotakes advantage of the principal component analysis method to extract effectivefeatures for oil seal defects classification. Meanwhile the project explores the supportvector machine classification and recognition method detailedly, builds classifiers foroil seal upper end area and lip area respectively, and fufills oil seal surface defectsclassification and identification.更多还原