- Cementitious materials are often characterized through the use of advanced analytical techniques to understand the macro-, micro-, and nano-scale properties, including phase formation during hydration, and subsequent potential deterioration mechanisms which can affect service life. A major limitation with using such analytical techniques to quantify solid phases in cementitious systems is that many techniques are destructive in nature. If one wants to monitor changes over time, samples must be extracted from different locations of the same sample, or from a different sample, at different times. This limitation can hinder the ability continually monitor the desired property. One method to extract quantifiable information non-destructively is x-ray computed tomography (x-ray CT). X-ray CT is a non-destructive, non-contact technique that uses computer-processed x-rays to produce three-dimensional tomographic images of specific zones of a sample. This technique has been successfully used in many different aspects of research, including medicine, geo-sciences, and materials science. The use of x-ray CT has been applied to cementitious systems but has been predominately limited to qualitative or semi-quantitative analysis. Quantification of cementitious properties has been hindered through two means, spatial resolution and low contrast between solid phases, including the unhydrated and hydrated phases. This low contrast has often led to quantification of void space within the cementitious sample. The work presented in this dissertation addresses methods to resolve the expected low contrast in x-ray CT images on cementitious and the potential for segmentation of the four main hydration products found in portland cement. This was achieved through an investigation of different image segmentation algorithms and a creative use of contrast agents to be bound into specific hydration products using a synchrotron x-ray CT. Advancements in x-ray CT optics and data collection are continually improving image resolution, therefore it is not discussed in this dissertation.
Published literature on the use of x-ray CT in cementitious materials often does not include a thorough description of the image processing procedures used for analysis. The use of arbitrary, histogram-based threshold values can lead to biased segmentation and misclassification of the voxels in the image volume. Presented in this dissertation is a method to deconstruct the greyscale values of a histogram into individual Gaussian curves in an unbiased manner. The greyscale values of laboratory synthesized calcium-silicate-hydrate (C-S-H), calcium hydroxide (CH), monosulfate (AFm), and ettringite (AFt) were determined to provide a baseline for threshold values. Pure phase, binary, and quaternary mixture samples of the four aforementioned phases were studied. A Gaussian probability density function was applied to each phase and proportioned to the known mass of each phase in the binary and quaternary mixtures. Intersections of the Gaussian curves was determined as the threshold value. Quantification of binary mixtures was successfully done with exception to C-S-H and AFm mixtures. Low contrast between the phases was observed leading to difficulties accurately quantifying such mixtures. Similar success was observed in quaternary mixtures of phases. However, difficulties in segmentation were compounded segmenting AFm, C-S-H, and CH in these mixtures.
One method to resolve low contrast is to incorporate the use of contrast agents. Success in the medical field, and other limited successes in geo-sciences, provided the motivation to determine methods to incorporate contrast agents into portland cement hydrates. Literature reports a myriad of ions which can be incorporated into the structure of C-S-H, AFm, and AFt through various mechanisms, including substitution and absorption. Due to the limitations in segmenting AFm and C-S-H in their pure form, investigations for incorporating contrast agents to improve segmentation was done. Iodine was selected as the contrast agent to be substituted for the sulfate ion in AFm. Dual energy scans above and below the absorption edge of iodine was done, and the use of image subtraction allowed for quick and accurate segmentation of the C-S-H and modified AFm phases. However, difficulties segmenting C-S-H and CH were observed in quaternary mixtures of AFt, CH, C-S-H, and a modified AFm using the Gaussian deconstruction method to determine threshold values for segmentation.
Lastly, due to the difficulties achieving consistent results during segmentation when using histogram-based threshold values, a study determining the feasibility of local segmentation algorithms was done on binary and quaternary mixtures of the four phases. These algorithms often result in more desirable results by accounting for the spatial arrangement of the greyscale values throughout the image volume. Two local segmentation algorithms, watershed and Bayesian Markov random fields, were compared to the Gaussian deconstruction method. Results indicated both local segmentation algorithms resulted in more accurate quantification of the four phases, thus providing promise for future applications to hydrating portland cement.