Tomographic Imaging of Flow Processes

X-ray computed tomography (CT) is a powerful metrology that allows for nondestructive imaging of the internal structure of objects. Improvements in X-ray sources, X-ray detectors and increasing computation power have paved the way for dynamic tomography, which enables a new range of experiments. Dynamic experiments are ubiquitous in research fields such as carbon storage, geothermal energy, multiphase flow, rock deformation and research related to the oil and gas industry. The dynamic nature of the experiments constrains the imaging process in that the imaging has to be sufficiently fast to capture it. This constraint can drastically compromise the image quality of tomographic reconstructions if either exposure time or the number of projections is reduced to accommodate the pace of the dynamic experiment. In this thesis, we have developed novel reconstruction algorithms for dynamic experiments to provide high-quality reconstructions despite substandard data. We designed the algorithms for experiments that allow for a high-quality scan of the experiment before any dynamics take place. All algorithms we developed are based on the Simultaneous Iterative Reconstruction Technique, and the best performing of these algorithms had two additions, which improved it. The first addition is we initialise a reconstruction for a given time step with a reconstruction of the previous time step. The first time step is initialised with a high-quality reconstruction of the static system. The second addition is that each pixel in the dynamic reconstructions is constrained such they can only take a single value or a range of values. This constraint is derived from a segmentation of the high-quality reconstruction. We tested the algorithms with simulated data of differing quality to gauge their performance under different conditions. We showed that this algorithm performs far better than conventional methods and allows for a substantial reduction in imaging time. We also examined multiple stopping rules for reconstruction algorithms that estimate the optimal point to terminate a reconstruction. Terminating a reconstruction before or after the optimal point will either result in a blurry or noisy reconstruction. All tested methods provided accurate estimates of the optimal stopping point for data with a moderate or small signal to noise ratio (SNR). Two of the methods do not require any knowledge of the noise level in the data, which makes them very practical. Finally, we performed a dynamic CT experiment to study the scaling formation process. We injected a carbon steel flow cell with water supersaturated with BaSO₄ for over 150 hours while imaging it with a CT scanner. We used a modified version of one of the reconstructions algorithms we developed to reconstruct the dynamic data to improve the SNR of the reconstructions. Quantitative analysis of the reconstruction allowed us to gain new insights into the scaling formation process. From this analysis, we show that scaling formation has three distinct growth phases. This thesis and the contributions within it shows the applicability of dynamic tomography for a wide range of fields and how it provides novel insight within these fields.