Analysis Of Computer Simulation Algorithm Of Cement Hydration

Bentz et al in NIST have proposed a three-dimensional cement hydration computer simulation model based on digital images. They use scanning electron microscopy (SEM) to collect backscattered electrons(BSE) images and X-ray images from cement paste's 2-D slices, and then a decision tree-based method is applied to segment phases which buildup cement paste from these 2-D images. When segmentation is completed, space distribution characters of each phase such as area fraction, perimeter fraction etc. can be easily calculated by using a pixel-count method. These data can then be used for the reconstruction of cement paste's initial three-dimensional microstructure. Then a series of cellular automata-like rules are used to operate on all pixels in the initial microstructure, and the hydration process of cement is simulated by this way. In this thesis, we start with the segmentation and process of cement paste's 2-D SEM images. On the basis of researching and analyzing Bentz et al's image segmentation method, we transform the image segmentation question to a question of image matching, and propose a new image segmentation algorithm which is based on correlation measurement from the point of view of digital image process. In this new algorithm, 2-D X-ray images are collected for Ca, Si, Al, Fe, S, K, Mg, and grey-level value of the same point in these images are normalized and arranged as a vector in a particular order. In this paper, the vector is named as character vector, for it can be used to characterize each point in cement paste's 2-D image. The character vectors of each phase in cement paste which are named datum mark are obtained by composition analysis and numerical experiment. For each point in cement paste's 2-D image, correlation measurements between this point's character vector and each datum mark are calculated, and the phase which has the maximum relativity with this point is assigned as this point's phase. So phase of each point in cement paste's 2-D image can be identified in this way, and the cement image is segmented. By comparison, we find that new algorithm can achieve the same segmentation precision as Bentz et al's algorithm, and can easily be extended too. In new algorithm, it is not necessary to calculate grey-level threshold for each X-ray image which is inevitable in Bentz el al's method. So the new algorithm is more general, computational and error tolerance.