Recently, techniques have been developed to accurately capture the detailed features of multi−phase cement particles in two dimensions (Bentz and Stutzman 1994). To obtain an image of an actual cement, the cement to be characterized is dispersed in a low-viscosity epoxy which is subsequently cured. The sample is polished and a 100nm thick coating of carbon is evaporated onto the polished surface to eliminate specimen charging in the scanning electron microscope (SEM). In the backscattered electron (BE) image, an example of which is shown in Figure 7, brightness is proportional to the average atomic number (Z) of a phase.
In the BE image, the ferrite phase appears brightest followed by the tricalcium silicate phase. Unfortunately, tricalcium aluminate, dicalcium silicate, and gypsum all exhibit similar brightnesses and thus cannot be identified on the basis of the BE image alone. To circumvent this problem, x−ray images are collected from the same area being viewed by the BE detector. Separate x−ray images are collected and thresholded for sulphur, calcium, iron, and aluminum as shown in Figure 8. By combining these four X−ray images with the original BE image, each of the four major clinker phases of a Portland cement powder along with gypsum may be distinguished. For instance, the presence of iron in the x−ray image indicates the ferrite phase while the presence of aluminum but not iron indicates the tricalcium aluminate phase. Similarly, the presence of sulphur indicates gypsum. Tricalcium silicate and dicalcium silicate are distinguished on the basis of brightness of the BE and calcium x−ray images. In cases where this separation is still difficult, a silicon x−ray image can be obtained and also used in the segmentation process. Finally, the segmented image is filtered using a median filter to sharpen the phase distinction in the final image and overcome some of the noise present in the BE and xray images. In the median filter, each pixel in the image is reassigned to be the phase occupied by the majority of its neighbours, excluding porosity, if this majority value exceeds a preset limit.
After combining the five images shown in Figures 7 and 8 and applying the mediantype filter, the final image of the cement particles (segmented into four clinker phases plus gypsum) shown in Figure 9 is obtained. This image can be used as input for a two−dimensional CA model. Accurately capturing the complexity of the starting cement material in this fashion greatly enhances the realism of the hydration model and allows it to be applied to complex real world problems requiring knowledge of the distribution of each phase in a specific cement. Examples include the study of sulphate attack, where sulphate ions infiltrate a cement−based material and undergo expansive reactions with only the aluminate and CH phases of the hydrated cement, and the diffusion and binding of chloride ions in a cement paste (Bentz and Garboczi 1993).