SIA OpenIR  > 装备制造技术研究室  > 学位论文
题名: 聚类分析理论研究及其在激光拼焊质量检测中的应用
其他题名: Theoretical Research of Clustering Analysis and Its Applications in Quality Inspection of Tailor Welded Blank
作者: 张承宁
导师: 赵明扬 ; 罗海波
分类号: TG665
关键词: 激光拼焊 ; 质量检测 ; k-means ; 聚类技术 ; 图像分割
页码: 115页
学位专业: 机械电子工程
学位类别: 博士
答辩日期: 2013-05-23
授予单位: 中国科学院沈阳自动化研究所
学位授予地点: 北京
作者部门: 装备制造技术研究室
中文摘要: 本文以中国科学院知识创新工程方向项目“全自动激光拼焊成套装备关键技术研究与示范应用”为依托,针对激光拼焊焊后焊缝表面质量视觉检测系统的关键技术进行了研究。论文总体上分为三个部分:第1部分是对聚类分析方法的研究。首先,对层次聚类理论进行了阐述,分析了层次聚类方法优缺点,提出一种基于链关系的层次聚类方法。该方法在小类内大类间分布假设下可以确定模式点类别数,并且避免了传统层次聚类迭代过程中的模式点选取随机性;其次,阐述并分析了划分聚类分支特点,提出基于空间模式点预测的k-means方法。该方法在两个方面进行了改进,一个是对空间分布进行了Hopkins系数的预测,可以确定分布类型是否适用于k-means方法;另外引入A.K.Jain的聚类有效性相对指标确定类别数。实验表明,该方法对于空间模式点具有更为准确的评估和聚类结果给出;最后,针对k-means方法局部最优问题,提出k-means全域类别建模方法。重点进行了类中心初始位置的相关讨论。由于k-means方法的局部最优问题的讨论往往只关注类别数设定的影响,而对于初始位置的影响并没有过多的涉及,所以有必要进行深入研究。m类n中心的类高斯模式点分布空间中,我们首先确定了k-means的聚类机制,明确1-NN始终是该方法遵循的规则。2类观点被引入目的是用来简化问题的讨论,并在此情况的基础上进一步强化条件,将类高斯分布特殊化为高斯匀质薄板。在k-means聚类机制与强化条件下,讨论了一种小概率事件和一种零概率事件,目的是明确位置因素导致的局部最优性的不可能形式。最后,在可能的局部最优形式下,我们给出了局部最优的古典概型表达式。第2部分是在完成聚类分析问题的研究后,从图像分割与聚类分析方面进行结合与应用;提出一种基于灰度级抽取的图像分割方法,将聚类技术作为直方图的重要补充来进行。基于直方图分割的前提是各像素级点集有聚团行为,且应对应像素级统计峰谷形态信息。这个前提其实是一个先验,是方法设计的假设。我们面临的分割问题,确切的说,是不具有这种确定先验的。本文以空间近邻关系信息的考察为准则,明确具有该近邻关系的像素级点集应该被聚合,进而提出了一种像素级抽取的方法。避免了传统直方图将顺序的像素级序列划归为一类的缺点,能够更为准确的进行图像分割。第3部分在实际应用方面,着眼于视觉检测的关键图像分割,指出常用方法的不足,给出多特征结合的图像分割方法。所做工作力求为我国激光拼焊检测装备的研制提供理论和技术支持。激光拼焊质量检测中,焊缝区域的识别的关键步骤是图像的分割。考虑到实际生产中可能遇到的油烟覆盖,热影响区不规则等因素,更加鲁棒的分割方法被提出。本文从焊缝物理形貌进行考察,确定了Laws纹理能量模板来进行区域的凸显。分块操作的目的是降低运算复杂度和描述的准确性,利用一致性、对比度、熵、均值、方差五个特征来进行每个子块的提取。最后利用主成分分析考察冗余特征,明确2特征空间下的焊缝聚类区域,进而完成了焊缝的分割。
英文摘要: The dissertation is divided into three main parts: the first part concentrates on clustering analysis. Details and critical issues are discussed, and improvement are given respectively; Firstly, hierachical clustering, compared with partional clustering, is good for exploring arbitrary shape of patterns. However, due to the continuous procedure of hierarchical clustering, the mechanism of cluster number confirmation is lack in this manner. This dissertation proposed a chain relation clustering method, which relies on the neighborhood growing fashion. The combined clusters are investigated in each iterative step, and the mean and variance of each cluster are calculated. The neighborhood growing is forbidden if there exist an extraordinary change in mean and variance. Noted that the chain relation clustering is under the hypothesis of grouping existence in pattern space, and there should be a long range of contant clustering number. Secondly, as for k-means, its implementation could follow the rule of Gaussian distribution of patterns in space. The prediction of patterns in the space is crucial before implementing k-means clustering. The clustering tendency for estimating the space is employed in this dissertation, as the foundation of using k-means. Additionally, k-means needs specified number of clusters, thus the optimal number of clusters should be confirmed. Here, the relative index of clustering validity is introduced to figure out the issue. Predition of pattern distribution and clusters number confirmation form the improved k-means method. Third, in the section of local convergence of k-means discussion, the attention is paid to initial center positions. Usually, the initial center number plays an important role of local optimal problem, and the position of initial centers is seldom discussed in literatures, the necessity of studies on this issue should be given. With m clusters and n centers Gaussian distributed patterns, we firstly explain the k-means iterative mechanism which is given as the 1-NN principle. 2-class viewpoint is introduced in order to simplify the problem, based on this, the strengthening condition is proposed where a supposed Gaussian thin sheet is the substitute for Gaussian distributed patterns. With the strengthened condition and k-means iteration mechanism, the discussion of little probability event and nill probability even is given. This is aimed at giving impossible local convergence form of ultimate clustering result. Finally, in possible form of local convergence, we propose the local optimal expression under classical model of probability. The second part focuses on the combination of clustering and image segmentation, which is followed the clustering analysis; Considering the combination of image segmentation and clustering analysis, the dissertation focuses on the improved histogram segmentation. Histogram based segmentation is under the grouping hypothesis, and the statistical form of corresponding gray level is also presented as peak valley characteristic. This hypothesis is actually a priority which determines the method design. However, the encountered segmentation issue, to be exactly, does not have that priority. So, the dissertation proposes the neighborhood estimation principle, this principle gives the rule that all special neighbors should be gathered, and these corresponding gray levels should be extracted to form the segmentation. Different from the conventional histogram based segmentation; there is no ordinal gathered gray levels in the manner of statistical meaning. The third part considers the practical application, drawbacks of conventional segmentations are figured out, and gives the multi-feature based image segmentation. Studies in this dissertation are provided for the development of the tailored blanks laser welding machine in our state. In quality inspection of laser welding, the recognition of welding seam is the critical step of image segmentation. Noted the practical producing procedure, oil cover and thermal effection dissortion are key factors to lead the failure of segmentation. More robust method should be proposed. This dissertation studies the form of welding seam, and confirms the Laws texture energy mask as the tools to highlight the interested region. Sub-regions are implemented considering the complexity of calculation and accuracy of description. Consistant, contrast, entropy, mean and variance features are used to describe each sub-region. Finally, PCA tool is employed to reduce five features to two features, and k-means is used in the feature space to confirm the welding seam region.
语种: 中文
产权排序: 1
内容类型: 学位论文
Appears in Collections:装备制造技术研究室_学位论文

Files in This Item:
File Name/ File Size Content Type Version Access License
聚类分析理论研究及其在激光拼焊质量检测中的应用.pdf(4774KB)----限制开放 联系获取全文
Recommend this item
Sava as my favorate item
Show this item's statistics
Export Endnote File
Google Scholar
Similar articles in Google Scholar
[张承宁]'s Articles
CSDL cross search
Similar articles in CSDL Cross Search
[张承宁]‘s Articles
Related Copyright Policies
Social Bookmarking
Add to CiteULike Add to Connotea Add to Add to Digg Add to Reddit
所有评论 (0)
内 容:
Email:  *
验证码:   刷新
标 题:
内 容:
Email:  *
验证码:   刷新

Items in IR are protected by copyright, with all rights reserved, unless otherwise indicated.



Valid XHTML 1.0!
Copyright © 2007-2016  中国科学院沈阳自动化研究所 - Feedback
Powered by CSpace