Microstructural Segmentation and Quantification of X-ray Tomographic Images of Concrete using Computer Vision

Concrete infrastructure is not as durable as desired, although concrete is the most consumed engineered material in the world by mass. To achieve carbon neutrality and slow down global warming, the construction community is under increasing pressure to improve the sustainability and durability of next-generation high-performance concrete.The desire to extend the life cycle of modern concrete infrastructure and reduce its lifespan carbon footprint has led to renewed interest in determining the reasons for the long-term durability of ancient Roman concrete. Besides, fiber-reinforced cementitious composites (FRCC), initially introduced in the 1990s, show the promise of controlling crack propagation, enhancing tensile ductility, and improving the structural durability of concrete infrastructure. The composition and microstructure of ancient Roman concrete and FRCC are the key factors of the long-term durability and ductility under mechanical loadings and environmental exposures. Understanding the microstructural mechanism of ancient Roman concrete and FRCC’s high performances in the mechanical tests (e.g., uniaxial compression test, uniaxial tension test, and three-point bending test) and durability test (e.g., water absorption test, alkali-silica reaction test, and self-healing test) is essential to design the next-generation concrete. There are three main gaps in the current state-of-the-art that need to be investigated: (1) mechanisms for the long-term durability of ancient Roman concrete in water; (2) mechanisms for the crack resistance capacity of ancient Roman concrete under loading; (3) mechanisms for the ductility and durability of FRCC. However, the previous studies' lack of in-situ analytical methods and volumetric representation for concrete’s microstructure under different field conditions limits the ability to cross the gaps.In this thesis, synchrotron X-ray computed microtomography (μCT) experiments were conducted to characterize the three-dimensional microstructure of ancient Roman concrete and FRCC to new information on the above-mentioned questions. Three sets of experiments were conducted to (1) characterize the microstructure and predict water permeability of ancient Roman concrete using μCT and neutron radiography; 2) characterize the microstructure evolution and ductile fracture pattern of ancient Roman concrete in the in-situ compression test and in-situ three-point bending test; and 3) characterize the microstructural evolutions of FRCC during the in-situ compression test, in-situ three-point bending test, and ex-situ durability tests. The state-of-the-art machine-learning derived pipeline for automated μCT image analysis was developed to process the three-dimensional raw μCT image, classify different components (phases), characterize the morphological properties, and conduct the statistical analysis. The pipeline outperformed the traditional methods on multiple tasks, especially image segmentation, in the accuracy, speed, robustness, and reproductivity. In the computer vision analysis, preprocessed μCT images were segmented into different phases accurately using machine learning-based and deep learning-based image segmentation models. The advanced algorithms and prototype software quantified the structure, distribution, orientation, and connectivity of the pores, cracks, and fibers in segmented phase images. For in-situ experiments, advanced digital volume correlation (DVC) algorithms calculated displacement map, strain field, strain localization, crack initiation, and crack propagation based on the 3D tomograms of the reference state and the deformed state. Besides, the results of other experiments (e.g., neutron radiography images, scanning electron microscope (SEM), and mercury intrusion porosimetry (MIP)) were combined in the data processing pipeline to investigate the sample microstructure systemically.The experimental results indicated that the highly porous ancient Roman concrete samples have similar water capillary penetration coefficients to modern PC paste/concrete samples. The low pore connectivity decreased the permeability and contributed long-term durability of ancient Roman concrete. Ductile fracture patterns were observed once cracks were introduced in ancient Roman concrete. The optimal aggregate grading curves and isolated pore structure contributed to the crack resistance. For crack resistance mechanisms in in-situ tests, the crack initiation happened at the ITZ around aggregates, and micro-cracks were diffracted and furcated by aggregates, especially the relict lime clasts in ancient Roman concrete samples. The reinforcement in the ITZ of aggregates obstructed crack propagation and produced toughening. The furcation and multi-crack propagation during the stable fracture pattern consumed additional energy and contributed to the ductility of ancient Roman concrete.Fibers play an important role in the ductility and durability of FRCC. Fibers reinforced the fracture planes in the FRCC through fiber debonding, bridging, bending, stretching, and fracturing under tensile load, which caused multi-crack propagation, reduced the average crack width and improved the toughness. During the uniaxial tensile test, the strain localization was observed around the connected pores with a volume exceeding 0.01 mm3, and multi-cracks propagated through these regions. Therefore, filling the pores within a specific size range using micro-scale fine aggregate may further improve the strength and ductility of the specimen. In durability tests, fibers controlled the average crack width and reduced the connectivity of the pore network, which contributed to the high durability of the FRCC. The work is important for providing mechanisms to design high-performance concrete at the microscale using a bottom-up approach