Research On Reconstruction And Parallel Of Cement Particle

Micro-structure of materials impacts the properties and characteristics. Gaining clear ideas quantitatively of the relationships between macroscopical performance and Micro-structure of materials is the primary researching objective of Computational Materials Science. For the composite of cement, several performance indicators, such as fluidity, condensation sclerosis, micro-structure, durability, are affected directly by hydration resistance. The properties of cement can be predicted more reliable by the exact simulation of the hydration processes. To simulate the hydration processes of cement, the first step is getting the initial 3D micro-structure that is three-dimensional distribution of cement particles before the hydration resistance.In recent years, with the fast development of the computer and graphics technology, the 3D micro-structure of cement particle can be built by 3D images and graphics simulation techniques. It provides a new scientific method and technological approach for the research and development of cement material.In this paper, 3D cement particles are reconstructed based on digital image processing technology. Beginning with 2D sectional images of cement particles, firstly, the accurate particle contours are extracted which is combined with the characteristic of cement particles. Then the 2D contours are registered. Registered contours are converted to 3D points with the "normal vector". Finally, adopting the reconstruction method based on scattered points, the 3D micro-structure of cement particles is built. At the same time, for increasing the reconstruction speed, we execute the parallel process of the reconstruction method. The major research contents are as follows:We studied the image contour extraction methods and analyzed the downsides of Snake method and level set method, which are the classical methods in contour extraction based on active contour models. Combining with the characteristic of sectional images of cement particles, fast global minimization of the active contour model was used to extract the raw contours. This method which combines the geodesic/geometric active contour (GAC) model based on level set method and the active contour without edges (ACWE) model based on Snake method by means of a weight function, can overcome the shortcomings of the above two methods and extract the right image contours. The contours of gray homogenization were obtained by means of the largest between-class variance method. Accurate particle contours were extracted by further processes.In image registration, there are two challenging problems: feature correspondence and transformation parameters estimation. To solve the two problems, in this paper, we developed the PBIL-MEP contour registration algorithm which need not build the registration model manually. Firstly, the feature points were obtained by mean of the centroid balance sampling method. The feature points were matched by the PBIL improved ICP. Then the registration model was established automatically using MEP which used the matched feature point set as the training set. Finally, the particle contours were registered in the registration model. The experiment results show that the PBIL-MEP method has the characteristic of simple realization, high executing efficiency and accurate registration. It is an effective and automatic image registration method.This thesis applied the MPU reconstruction method which is based on 3D space points to reconstruct cement particle surface. The registered contours were converted to 3D points with the "normal vector" adopting the plane fitting method. For cement particles, the MPU method has the weak point that the reconstruction surface is not smooth. We employed the orthogonal polynomials as local fitting function to increase the performance of the traditional MPU method. The improved MPU method reconstructed the idealized 3D implicit surface.To accelerate the reconstruction speed, we developed the parallel mechanism of MPU method. The algorithm adopted the hybrid parallel mode of Master-Slave and Single Program Multiple Data and introduced multithread to control load balance. With the parallel mechanism, the execution efficiency of MPU was improved and the execution time was reduced.