【作者】 王静；

【导师】 张树生；

【作者基本信息】 西北工业大学 ， 航空宇航制造工程， 2004， 博士

【摘要】 图像二、三维重建与分析技术,可以使人们在不损坏被测对象的前提下,借助穿越物体的X射线等投影数据获得其内部信息,建立物体的二维断面图像及三维模型,并进一步实现对其内部结构的分析和理解。该技术最早出现于医疗诊断领域,是CT(Computerized Tomography,计算机断层摄影)及MRI(Magnetic Resonance Imaging,核磁共振成像)等设备的技术基础,目前更广泛应用到工业产品反求及无损检测、地质勘探、文物考古等各个领域。 工业应用领域研究的复杂产品多属装配体,具有大体积或多分支的特征,因此在基于工业CT的复杂产品图像重建与分析过程中,面临着大体积产品的切片获取、分叉结构处理、复杂配合关系分割等关键问题;加之工业产品本身复杂多样,无法获取类似医学上人体模板的统一模型,使得面向复杂产品的工业图像重建研究充满挑战。本文针对航空航天领域复杂装配体或大尺寸产品无损检测及产品反求的需要,讨论其中的关键环节和阻碍这项技术进一步工程化的难点。论文的主要研究内容和成果如下: 1.二维CT图像重建及仿真 实现面向装配体的投影数据仿真生成:传统的ICT仿真大都是针对单个零件的,论文针对航空航天领域大量的装配体产品,分析其“多零件、多材质”的特点,采用面向对象的仿真建模思想和方法对ICT系统进行系统仿真建模,有效而完整地实现了装配体建模,最后实现了专门面向装配体产品的投影仿真系统,可以获得任意扫描方式、任意角度的投影图像及数据文件,方便地为图像重建与分析提供丰富的基础研究数据,为后续章节的研究奠定基础。 提出并实现不完全投影条件下的切片图像重建算法:针对固体火箭发动机等大体积产品无法实现完全投影的问题,论文首先分析切片图像的重建原理,并根据不完全投影的产生原因,将其分解为中空型、稀疏视角型和限制视角型三类,针对不同的不完全投影类型,分别提出射线补充法、两层级联神经网络法、遗传重建算法三种算法,实现了各种投影条件下,由不完全投影到切片图像的重建,最后分析算法的重建效果及效率,提出复合型不完全投影的重建方法,从而实现不完全投影情况下的二维CT图像获取。 2.基于CT切片的三维重建与结构分析 提出并实现面向目标零件的装配体分解及零件间装配关系识别方法:针对装配体切片重构过程中存在的各零件相互干扰而难以直接进行匹配重构的问题,论文提出进行装配体零件分解的思路,首先建立灰度轮廓树及轮廓森林,并通过灰度阈值及连通体生长与聚合分解出各个组成零件,在零件分解的基础上建立各零件的轮廓森林,为零件的三维重构做好准备,并进一步利用轮更多还原

【Abstract】 In virtue of 2D and 3D image reconstruction from X-ray projections, people can get the interior information of an object in a non-destructive way and carry out the inner structure interpretation and analysis. Appeared firstly in the medicine field, 2D and 3D image reconstruction technology is the basic of the modern instruments such as CT and MRI and is widely applied in various domains including reverse engineering, non-destructive test, geology exploration and archaeology.However, there are still many key problems unsolved, especially in the ICT based industrial image reconstruction aspect because of some characteristic of industrial products such as the complex part structure, huge volume and multiple styles etc. Aiming at the non-destructive test and reverse engineering on complex assemblies and huge products in the field of aviation and astronomy, this thesis discusses some key and difficult issuses in 2D and 3D image reconstruction process to promote the development of CT technology. The main research contents and contributions in the thesis are as follows:Projection simulation for assembly products: Simulation technology is introduced to the image reconstruction and an assembly-oriented projection simulation system is developed to provide abundant research data for following image reconstruction. Different from traditional ICT simulation systems which are single part oriented, we studied the characteristics of multi-part and multi-material case for ICT simulation of assembly product, and realize system modeling by using objected-oriented modeling method. At last, the assembly is well modeled and the simulation system is realized for assemblies, projection data can be got by arbitrary scan style and in arbitrary projection angle.Slice image reconstruction from incomplete projections: Slice reconstruction theory is analyzed firstly, and three kinds of incomplete projection, hollow in center region and sparse viewing angle and limited projection angle range, are considered respectively according to the projection style. The projection rays supplement approach, joint two-grade neural network approach and genetic reconstruction algorithm are proposed respectively for above three cases to realize slice reconstruction from incomplete projections. The reconstruction effect and efficiency are analyzed, and the reconstruction for compound incomplete projection case is also studied.Parts-oriented decomposing and assembly relation recognition for assemblies: 3D reconstruction from slices for an assembly is a difficult task because the parts in it interfere with each other and the contours in adjacent slices can’t bematched and tiled directly. This thesis studies this issues in detail. Firstly the contour tree with gray level for a slice and the contour forest for whole slices are established, and then based of them, all the parts are decomposed and contour forests for each part are set up to prepare for part reconstruction, at last the assembly relations between parts are recognized through contour tree and constructional relation figure is generated.> Part’s 3D surface reconstruction from slices based on the part slice tree: Aiming at the matching problem caused by multi-branch, distortion or high complexity of the industrial product, we proposed a series of algorithms including global matching using ellipse parameter, local matching for contour segment, branch disposal through slice projection and post-matching based on area relations to implement matching between slices and part’s 3D reconstruction.> Knowledge based 3D-feture recognition and representation: one of the differences between industrial part and human organ is that the former often contains a lot of 3D geometry features and could be described in total. So some common features and their corresponding slice contour as well as contour series are summarized as knowledge in the thesis, and a knowledge base is set up. By knowledge-based feature recognition, the relationships among all features are analyzed and a feature tree for the part is formed. To facilitate the data exchange among different applying systems, XML is utilized to represent parts instead of traditional triangle surface representation method.> Prototype system design for 2D and 3D industrial image reconstruction and analyze: By using dynamic library technology and VTK (visualization tool kit), a prototype system named ICT-SliceVolume is designed and implanted, the system integrates not only most research results but also some necessary algorithms for industrial image reconstruction to validate the key technologies proposed in this thesis.This research work is sponsored by the two funds of China Aeronautical Foundation Science Funds& Astronautical Innovation Funds (99H53111 and 2001 [181]), and it reflects the author’s main research results since Mar. 2000.更多还原