【作者】 方黎勇；

【导师】 李柏林；

【作者基本信息】 西南交通大学 ， 机械设计及理论， 2009， 博士

【摘要】 随着科学技术的不断进步,对零件的质量和可靠性的要求越来越高,而难以被直观检测和定量分析的零件内部缺陷对其具有关键的影响。目前零件内部缺陷检测主要集中在二维阶段,为了深入地研究零件内部缺陷的形状以及在外力或应力条件下的变形、拓展等演变情况,从而对零件的可靠性、寿命等进行判断提供确实可信的理论依据,就有必要对零件内部缺陷进行三维重构。本文采用了目前无损检测界公认的一种最优检测手段——ICT(Industry Computerized Tomography)技术,对零件内部缺陷进行检测,从而获得该零件的序列切片图像数据。然后,借鉴了产品反求工程思想,结合零件内部缺陷特点,对零件内部缺陷的三维重构关键技术进行研究。首先,归纳了零件内部缺陷的ICT切片图像特点,分析了现有的图像缺陷分割算法,提出一种基于自适应形态学滤波的缺陷目标分割方法,即利用分水岭变换对图像的三维灰度分布曲面进行处理,获得形态学变换的结构元素的初始模板,再对其进行椭圆拟合作为形态学变换的结构元素,模拟出背景图像,与原始图像进行差运算获得缺陷目标。其次,研究了缺陷像素轮廓点的矢量化和图像坐标与三维空间的坐标转换,分析了同属于一缺陷体的不同层切片上轮廓的相似性特点,引入了相似测度的概念来定量描述相邻层间缺陷轮廓的相似程度,从几何和非几何信息来计算其值。采用了缺陷轮廓的拟合椭圆参数来定量描述不规则缺陷轮廓的几何信息,而非几何信息则采用缺陷轮廓内部的像素灰度来描述。以相似测度值为依据,提出了层间缺陷轮廓的匹配算法。采用了整体匹配和局部匹配的匹配策略来处理分叉的识别和判断。为保证匹配的准确性和减少运算量,提出了一种基于缺陷轮廓像素点的空间邻域生长的候选匹配区域确定方法。再根据三维拓扑重建的结果,研究了真伪缺陷的识别方法,给出了判断准则。通过实例验证了所提算法的准确性和可靠性。然后,分析了相邻层上相匹配的缺陷轮廓特点,提出了采用一种三角网格的拼接方法,即采用Freeman链码差函数快速检测缺陷轮廓的特征点,然后对轮廓数据进行精简,根据相似性理论来处理相匹配轮廓的特征点对应,提出了相应准则对其对应结果进行增补使其达到一一对应,再采用局部最优的三角网格拼接原则进行网格拼接。该方法在保证了细节特征的同时有效地减少了数据量。最后,借鉴了现有的三维重建软件的特点,在上述研究的基础上,设计开发了针对序列ICT切片图像的缺陷三维重构原型系统,并对重构的缺陷三维模型主要参数进行了研究,得出了它们的计算方法。更多还原

【Abstract】 With the advancement in science and technology, there are increasing demands on quality and reliability of mechanical parts. Internal defects in mechanical part have important influence on overall quality and reliability, but they are difficult to detect directly and analyze quantificationally. Conventional practices are concentrated on 2-D images. To fulfill the needs of research on defects configuration and deformation under forces or stresses, furthermore to provide accurate and reliable theoretical basis for judgment quality and reliability of mechanical parts,3-D reconstruction of internal defect in mechanical part is systematically studied in this dissertation. ICT(Industry Computerized Tomography) is referred as the best technique in Nondestructive Testing field, so it is used to detect and obtain the slice images of internal defects in mechanical part. This dissertation focuses on the key technologies of internal defects 3-D reconstruction in mechanical part based on its ICT slice images.Firstly, characteristics of ICT slice images of internal defects in mechanical part are summarized and advantages and disadvantages of existed defects segmentation methods based on images are analyzed. Based on previous research, an adaptive morphological filtering method is presented to extract all potential internal defects in mechanical parts. In this method, simulation background image is obtained from morphological operation to original slice image, and the structure element is set as the fitting ellipse of the maximum region contour after watershed transform of original slice image. Therefore, the defects are well segmented out by subtraction operation between simulation background image and original slice image.Secondly, vectorization of defect contours data and coordinate transform between image and 3-D are studied after 3-D coordinate system is constructed. The similarities of defect contours on between different slice images belong’to one defect body are summarized. The conception of similarity measure is introduced to describe the similarity degree between two defect contours on neighboring slices, and the geometry and non-geometry information are considered to calculate the value of similarity measure. The parameters of fitting ellipse of defect contour are used to describe geometry information, and the pixel gray information internal defect contour is used to describe the non-geometry imformation. A matching algorithm of defect contours based on similarity measure is presented, and a matching strategy of two steps is used to deal with the branching of defect contours. For improving accuracy rate and reducing computational complexity of contours matching, a method of determination candidate matching region based on one order space neighborhood of contour points is presented. Adjustment criterion to distinguishing the true or false defects from all potential defects is proposed according to the results of 3-D topological reconstruction. The accuracy and reliability of presented methods are validated by experiments.Thirdly, a method of triangular mesh building is presented based on characteristics of defect contours data from serial ICT slice images. In this method, the first step is feature points detection and contour data reduction by Freeman code difference function, the second step is feature points matching based on similar theory of feature points, the third step is feature points supplement to deal with the instance of feature points with no matching relation, the last step is triangular facet construction based on a local optimal method. By this method, not only detailed characteristics of defects are kept but also data amount is reduced.Lastly, a prototype system is designed and developed based on the foregoing researches. Functions and structure of the system are designed in this prototype system, and calculation methods of some key parameters of reconstructed defects model are presented.更多还原