【作者】 杨文丽；

【导师】 王乘； 曾致远；

【作者基本信息】 华中科技大学 ， 系统分析与集成， 2012， 博士

【摘要】 随着石渣坝、高土石坝、公路铁路、高层建筑物等工程的不断兴建，粗粒土作为主要的填筑材料，其应用也越来越广泛，因此粗粒土组构的力学特性也越来越引起坝工界的重视。为适应石渣坝等土工建筑物的应力和变形统计分析的需要，粗粒土在压力状态下的应力-形变关系成为岩土力学领域重要的研究内容之一，而粗粒土的微观组构特性是其应力-形变关系研究的基础。粗粒土组构（micro-fabirc)是影响粗粒土的工程性质的关键因素，对粗粒土微观组构图像的分析和处理也是粗粒土组构研究由定性走向定量的关键技术。针对目前基于粗粒土微观组构的研究基本都是基于二维图像处理，由于真实存在的粗粒土颗粒是基于三维空间分布的，因此二维研究方式存在一定的局限性，不够精确和客观，因此结合粗粒土图像的特点和实际的应用需求，本文首先利用计算机CT技术获得序列断层粗粒土微观组构图像，然后运用序列断层图像三维重建及可视化技术重构二维粗粒土微观组构图像的三维模型，进而对三维粗粒土颗粒模型进行微观组构要素提取和分析，为建立粗粒土组构力学模型奠定了坚实的基础。基于以上的研究问题与技术思路，本文依托"湖北省长江科学院高应力及复杂加载条件下土石料工程特性试验研究"项目开展了具体的研究工作，并取得了有实际工程应用价值的部分研究成果。论文首先详细总结和分析了目前粗粒土微观组构的国内外研究现状和方法，给出了本文研究方法的依据和理由，全面介绍了本文针对序列断层粗粒土图像三维重建及可视化技术研究的相关理论基础，并对三维重建及可视化技术的关键研究进行分析和总结，提出本文的研究技术路线和具体研究内容。其次论文对序列粗粒土图像三维重建及可视化技术研究中的序列图像分割技术进行分析和研究，结合粗粒土图像的具体特点，首先对粗粒土图像进行噪声滤波和插值等预处理一，得到较高质量和连续完整的序列断层粗粒土图像，进而提出种基于快速聚类的分水岭分割算法对粗粒土图像中粘连颗粒进行快速有效地分割，并针对部分颗粒过分割情况进行聚类处理，从而保证批量粗粒土图像颗粒分割的精确性，并根据实际的应用需求提出一种不同压力控制下粗粒土图像的目标颗粒序列二维图像提取方法，通过观测和分析不同压力状态下目标颗粒的形态特征和位移状态，为后续粗粒土组构力学效应分析和建立组构力学模型提供了很好的图像数据支持。再次，论文以序列断层粗粒土图像分割技术为基础，提出序列断层粗粒土图像的三维重建及虚拟切割算法，首先介绍和分析了序列断层图像三维重建的基本流程和常见算法，并结合序列断层粗粒土图像的特点和实际的应用需求，提出一种基于数据分类的快速混合绘制算法，首先利用分割技术将重建数据进行分类，然后针对不同的分类数据的重要程度，分别采用不同的绘制算法对不同的数据进行绘制，从而既能保证绘制图像的质量，又能加快绘制速度。最后针对重建的粗粒土颗粒三维模型，进一步提出粗粒土微观组构要素提取方法和重建模型简化技术。本文在传统针对二维粗粒土微观组构要素提取的研究基础上进行三维空间扩展，对重建后的粗粒土颗粒三维模型进行特征提取和分析，包括粗粒土颗粒的三维几何形态和三维空间分布特征，并在不同应力状态下对微观组构要素变量进行分析，为今后进一步粗粒土颗粒空间力学模型的建立提供依据。另外由于三维重建后的粗粒土颗粒具备较多的细节信息，虽然使得三维模型精度较高，但是针对大规模数据集的三维重建，会导致三维表面模型包含大量的三角面片，成为了可视化和后续数据处理的严重瓶颈，因此本文为尽可能的减少重建三维模型中的冗余数据，提出一种基于粗粒土颗粒三维中心长轴提取和非均匀B-样条拟合的三维模型简化算法，该算法能够使用最少的数据抽象信息构造完整和光滑的模型，从而实现大规模数据集实时绘制、存储和处理。论文最后对本文的研究成果进行总结和展望，对论文中提出的序列断层粗粒土三维重建及可视化技术研究方法进行分析，并对后续有待继续研究的一些问题进行展望。更多还原

【Abstract】 With the continual construction of the slag dam, high embankment dam, highway andrailway, high building and other projects, granular soil as the main filling mateiral, has anincreasingly wide application and it is also more and more attractive to the dam sectors. Inorder to meet the needs of the stress and deformation analysis of the slag dam and otherstructures, the intensity and stress-strain relationship of granular soil under high pressureor other complex stress state becomes an important research content of rock mechanics.And the micro-fabirc characteristic of granular soil is the basis of the research forstress-strain relations. The micro-fabric of granular soil is one of the key factors whichaffects the engineering characteirstics, and the computer processing of the image ofgranular soil is also the key technology which make the micro-fabric research from thequalitative to the quantitative. The current research of the micro-fabric of granular soil ismainly based on2D image processing, while the actual granular soil is in3D space, so ithas some limitations in accuracy and objectives of2D method. According to the featuresof granular soil image, as to satisfy the practical needs of application, this thesis tires touse the CT technique to obtain the image sequences of granular soil, then reconstruct the3D model based on the2D granular soil images with3D reconstruction and visualizationtechnology, and also extract and analyze the features of granular fabric through the3Dmodel, which lays the foundation for the next stage research for establishing themechanical model of granular soil fabric. Focusing on the descirbed above problems andideas, the thesis, supported by Yangtze River Academy of Science of Hubei province, hasdeveloped the specific study and obtained many achievements with practical applicationvalue.Firstly, the thesis lists the research status of the micro-fabirc of granular soil, andpresents the reason and basis of our study method and introduces the related concepts andtheoires of3D reconstruction technology based on slicing image sequence, then also makethe analysis and summary of the key studies of3D reconstruction and visualizationtechnology, then leads to the specific content and method of the whole study.Secondly, This article studies and analyzes the slicing image segmentation technology,based on the features of granular soil image, atfer some pre-processing like ifltering andinterpolating, we can get the slicing granular soil image sequence with higher quality andcontinuity, then presents a new method with the watershed segmentation algorithm basedon fast clustering to accomplish the effective segmentation of linking particles of granularsoil in the image, here we adopt respective measures to deal with the over segmentation ofpatricles, which guarantee the accuracy of the segmentation, then also proposed a methodto extract the2D image sequence of a target particle rfom the slicing image of granular soil based on the different pressure conditions. The observation and analysis of thegeometric shape and motion-state of target particles under different pressure conditionswill become the considerable image data support to the following analysis and buildingthe micro-fabric mechanical model.Thirdly, the thesis proposed the algorithm of3D reconstruction and virtual cuttingbased on the segmentation technology of granular soil slicing image sequence. This partstarts with the basic lfow of3D reconstruction, and does the analysis and summary of thecommon algorithm of3D reconstruction, then presents a fast ray casting volume rendeirngalgorithm based on the data classification. The algorithm first needs to classify the originalreconstruction data using the presented segmentation method then make the differentrendeirng algorithm according to the different data. The choose of rendeirng detaileddegree is based on the signiifcance of data classification, which will not only guaranteesthe quality of the rendeirng3D model, but also increase the rendeirng speed.Lastly, the thesis further proposed the method to extract the micro-fabric elementsand the surface simpliifcation of the reconstructed3D model. The thesis gives anextensive study on the traditional2D extraction methods of the micro-fabirc, that is makethe research and analysis in3D space, including3D geometric shape extraction and thespatial distirbution features of granular soil particles under the different pressure condition,which will make a better data and basis for the further research of building the mechanicalmodel of granular soil particles in3D space. Then because of the3D model atfer thereconstruction has more detail information and make the data very huge, the visualizationof such data sets is already a bottleneck and it will become impossible unless newmethodologies are developed. Therefore, it is necessary to develop computer graphicsrendeirng techniques that use data abstraction and allow for different levels of detail, thethesis also presents a surface simpliifcation method based on the3D center long-axis andnon-uniform B-spline fitting algorithm to reduce the redundant data. The proposedalgorithm could give a smoother surface approximation representation, requires lessabstraction time, and uses a smaller descirption for data abstraction.Finally, the research achievements are summarized and directions for further researchare pointed out.In the end, this thesis summarizes and expects the above research results, analyze theproposed3D reconstruction and visualization technology of slicing image sequence ofgranular soil, and birelfy discussed the ifirther research of the thesis.更多还原