Brillouin scattering–based distributed fiber optic sensing (Brillouin-DFOS) technology is widely used in the long-term health monitoring of large-scale structures to provide early warnings of structural degradation for timely maintenance and renewal. Material cracking is one of the key mechanisms contributing to structural failure. However, the conventional strain measurement using a Brillouin-DFOS system (Brillouin optical time-domain analysis/reflectometry (BOTDA/R)) has a decimeter-order spatial resolution, making it difficult to measure the highly localized strain generated by a sub-millimeter crack.

This research develops and implements two crack analysis methods based on Brillouin scattering spectrum (BSS) data to overcome the spatial resolution–induced crack measurement limitation of the BOTDA/R system. The first method extracts the maximum strain within the spatial resolution around the measurement points by taking the derivative of the BSS data and tracking their local minimums. The crack can be located by comparing the variation of the extracted maximum strain within the spatial resolution around different measurement points along the fiber. The performance of the method is demonstrated and verified by monitoring the opening of a synthetic crack between two wood boards in a laboratory test. In the test, the crack width estimation error is $\pm$0.15 mm for a crack as narrow as 0.23 mm. The method is also applied to a thin bonded concrete overlay of asphalt pavement experiment, in which the growth of a transverse joint penetrating the concrete–asphalt interface is monitored. The method successfully distinguishes the position, traces the strain variation, and estimates the width of the sub-millimeter crack using a Brillouin-DFOS system with a 75-cm spatial resolution.

The second method back-calculates the strain profile by decomposing the BSS data. The method uses the least squares–based algorithm, non-negative least squares, to back-calculate the strain profile within the spatial resolution around each measurement point. The performance of this method is verified using the same wood board separation test as the first method. With the strain profile being calculated accurately, the crack width estimation error is improved to $\pm$0.005 mm for a crack as narrow as 0.23 mm compared to $\pm$0.15 mm error of method I. The method is also applied to monitor crack openings within a slag cement-cement-bentonite (SCCB) beam under four-point bending. The method can identify not only the crack generation but also the occurrence of slippage from the estimated strain profile. It successfully measures a sub-millimeter crack in the SCCB beam with an error of $\pm$0.005 mm.

One of the advantages of the DFOS technology is that it is designed for multi-purpose applications; that is, a fiber instrumented in the structure and connected to different types of interrogators can be used to take many types of measurements, such as strain, temperature and vibration. In this research, a fiber embedded in pavement is connected to an optical frequency domain reflectometry (OFDR) system, and DFOS technology is applied to detect and monitor the movement of vehicles by measuring their induced dynamic deformation of the pavement. From the footprint formed by the distributed dynamic strain profiles measured by the OFDR system, the moving direction and speed of the vehicles are accurately evaluated. Different vehicles and pedestrians are distinguished from each other based on their induced strain values and footprint patterns.

This research therefore develops two BSS-based methods to improve the performance of a BOTDA/R system in measuring the heavily concentrated deformation induced by cracks and investigates the ability of DFOS technology to detect and track vehicles.

Hydraulic fracturing is a technique for extracting unconventional resources by injecting high pressure fluid to crack the reservoir layer. This is a multi-scale and multi-physics problem that involves the behavior of rock matrix and injected fluid flow, as well as their interaction and resulting fracture propagation. The heterogeneities in the reservoir rocks make the fracture propagation a complex phenomenon and hydraulic fracturing an even more challenging problem.

In this study, a C++ based three dimensional Lattice Element Method (LEM) simulator is developed, which is capable of simulating both mechanical and fluid induced fracturing behavior in heterogeneous media. In this model, the mechanical response of solid is represented with a 3D lattice structure. The Timoshenko beam with embedded discontinuity is used as a solid lattice model, and exponential softening law is incorporated. The fracture surface is represented by the discontinuity generated in the lattice once it yields. While the fluid lattice network is a pipe flow system generated at fracture surfaces. This fluid lattice model follows Darcy's law and the principle of mass conservation, and interacts with solid lattice network through hydrostatic pressure. The 3D multi-physics problem is simplified into a network composed of 1D beam and pipe lattice, which can be simulated with a relatively low computational cost. The validations and comparisons between numerical and experimental results indicate that LEM is a promising tool for investigating the process of fracture development. Its simplicity makes it capable of simulating branching and complex interactions between multiple fractures, which is extremely difficult in continuum-based methods. A nonlinear softening law is implemented in the solid lattice model, which allows the simulation of the failure of materials with different degrees of ductility. In addition, the potential application of LEM to early crack detection is explored.

The model is used to investigate the influence of rock heterogeneities on the hydraulic fracturing process, especially the fracture pattern. The results show that, all the factors, including in situ stress field, scale of weakness and statistical distribution of properties, affect the results to a different degree. The nodal arrangement is one of the main factors that affect potential fracture path. The interactions between hydraulic fractures and pre-existing natural joints are also investigated. The hydraulic fracture penetrates into a joint when the normal stress on the joint surface is too small to stop the joint from opening under fluid pressure. Whereas a fracture crossing is more likely to occur when there is little or no slippage between two sides of the joints. This implies a possible mechanism of fracture crossing in which the fracture reinitiates on the other side of the joint due to tensile stress generated by the friction applied on the joint surface.

Earthquakes are simulated in the laboratory using a homogeneous rock analog material, taking advantage of highly controlled conditions to investigate the source properties of asperity rupture. The frictional interface consists of two transparent Poly(methyl-methacrylate) (PMMA) blocks with self-similar roughness profiles imposed by sand blasting and laser engraving to match our understanding of rock-rock interfaces. The resulting interface is imagedby a pressure sensitive film revealing the real contact area spread over sparsely distributed 1 mm patches called asperities. Stick slip events on the direct shear fault are recorded by an array of absolutely calibrated, single component displacement acoustic emissions (AE) sensors. Rupture events which are fully contained within the fault surface have radiation patterns consistent with double couple source mechanisms aligned with the fault plane and strike, making them earthquake-like events at the millimeter scale. A catalog is developed containing 50 Mw-7.5 to Mw-6.5 events which exhibit self-similar scaling properties in estimated magnitude and rupture area, which connects these events to previously observed earthquake scaling behavior at laboratory and natural scales.

The acoustic emissions sensors are absolutely calibrated by a method combining well characterized, broadband calibration sources with a Green’s function model for wave propagationthrough the PMMA base plate. The Green’s functions are carefully tuned for the samplespecific material properties of the plate, including anelastic attenuation. Each sensor is individually calibrated and found to have a sensitivity on the order of 1 V/nm. Displacement records for the experimental stick-slip events have sudden onsets, with first arriving pulses around 0.01 - 1 nm in amplitude and 1 - 4 μs in duration. The arrival pulses are often complex in shape, such as having two peaks rather than one smooth pulse. Displacement arrival pulses are directly proportional to the moment-rate function which describes the distribution of slip during the event source. Furthermore, the magnitude analysis of the events indicates a rupture area at the appropriate scale but generally larger than any one asperity observed on the fault surface. Together, these observations indicate that rupture events recorded likely have coseismic slip spreading over multiple nearby asperities.

In order to study the complex, multiple-asperity sources, a finite source inversion methodfollowing Hartzell and Heaton [1] is developed. The method is tested on forward modeled synthetic asperity ruptures, beginning with simple ruptures on single, circular asperity models. Although a brief attempt is made at applying the method to recorded experimental data, more testing and sensitivity analysis is required before the method may be reliably used. Continued work in laboratory scale seismic inversions shows great promise in providing a highly controlled environment for detailed source study of earthquake-like events.

Cities and people who live in them require clean water to thrive. However, the conditions of water infrastructure in the United States are concerning. Many water distribution networks (WDNs) are serving beyond their intended design life. The aging water infrastructure is reflected by the frequent water main break events across the country in recent years. Using data mining and natural language processing (NLP) techniques, this study validates the commonly-hold opinions that water main breaks cause severe societal troubles, including repair cost, local traffic disturbance, and water quality related health issues.

Hazard events often damage the already in-risk aging WDNs. To estimate the hazard impact (e.g. earthquake) to a WDN, this study developed a WDN hydraulics simulator, HydrauSim, which can quantify the WDN hydraulics (e.g., flow rate, etc.) under normal or damaged states. HydrauSim is highly optimized to be computationally efficient, making it feasible for large-scale networks and tasks that require repeated simulation runs (e.g., Monte Carlo simulation). Using HydrauSim, the post-earthquake response of a WDN in the San Francisco Bay area is simulated. The simulation was performed on East Bay Municipal Utility District's (EBMUD) main gravity feed zone. Around 200--800 pipes were estimated to break during the simulated earthquake events. On average, 25\% of demand nodes may experience insufficient water pressure levels, which can rise to 78\% for the worst-case scenario.

In real-life situations, failed pipes need to be isolated from the main network by closing the corresponding isolation valves to prevent the effects of individual events from spreading throughout the system. However, most utilities do not have sufficient valves installed, and the installed ones may malfunction at the time of usage. This study proposed an analysis framework for WDN pipe isolation risk considering valve condition uncertainties. It is found that the magnitude of the risk depends on the mean and variance of isolation segment sizes and demand distribution across the network.

Using dynamic programming, an optimal valve placement algorithm is developed to find the best place to install isolation valves to minimize the system risk. The proposed method is tested on two real-life WDNs. Comparing to the existing valve placement configuration, the proposed configuration significantly reduces the pipe isolation risk of the system. Furthermore, the proposed valve placement strategy produces a more robust network than the original one regarding valve failure scenarios. Pipe isolation risks are significantly reduced at all tested failure rates, and the risk-increasing trend (as the valve failure rate increases) is effectively restrained.

Due to resource constraints, it is impractical for water utilities to maintain all the isolation valves in a system. This study proposes a method to rank the isolation valves based on their potential failure consequences. The valve ranking algorithm utilizes network analysis methods and machine learning techniques to label valve maintenance priorities automatically. Simulation on real-life WDNs shows that applying the proposed valve maintenance strategy effectively reduces both the direct and indirect risk for the tested networks, especially under high valve failure rate cases.

This dissertation presents a development of an integrated framework of artificial intelligence (AI) systems for earth pressure balance (EPB) tunnel boring machine (TBM) tunneling. The framework is constructed based on the feedback loop control system. The AI systems are developed using machine learning algorithms and structured to follow the human cognitive model, i.e., sensing, perceiving, and decision-making. The development of the systems is conducted in three parts.

The first part discusses the characteristics of TBM data and the effects of data preparation on data-driven models. This is achieved by (i) proposing a knowledge-based EPB TBM feature taxonomy and (ii) investigating the effects of data aggregation and feature selection on prediction models. The investigation shows that models developed using different data aggregation levels produce comparable prediction trends and similar feature importance rank in the conditions of sufficient observations and predictors. However, models with a coarser aggregation level may appreciate higher prediction performance due to the lower variance. The investigation also shows that the knowledge-guided TBM feature selection offers benefits over embedded machine learning-based feature selections. The developed model can produce relatively consistent feature importance in different tunneling cases, indicating better generalizability of the model.

The second part proposes AI systems that perceive tunneling environments in real-time based on the streams of sensor data during tunneling operation. This is achieved by developing (i) a supervised AI system to interpret the encountered geologic conditions, (ii) an unsupervised AI system to detect the encountered geologic anomalies, and (iii) a supervised AI system that connects TBM data to the ground monitoring data and estimates tunneling-induced ground movements. The proposed geologic interpretation system uses either Random Forests (RF) classification or regression algorithms to infer the geologic transitions along the tunnel alignment. The proposed geologic anomaly detection system combines Principal Component Analysis (PCA) to project the data into a lower dimension space and Local Outlier Factor (LOF) to measure the degree of the anomaly of the projected data points. The proposed tunneling-induced ground movement estimation system uses RF regression to approximate any shape of ground movements solely based on TBM operation data and tunnel spatial geometries without prior assumptions on the ground movement shape, geologic material parameters, and the expected ground loss.

The third part proposes AI systems that model tunneling and its decision-making processes. This is achieved by introducing (i) a combination of probabilistic graph modeling and structure learning algorithms as a tool to systematically explore the causal effect interactions contained in TBM operation data and (ii) a multi-output supervised AI system to determine multiple steering control parameters simultaneously and steer the TBM along the tunnel alignment. The study shows that Bayesian Network Structure Learning (BNSL) can potentially be used to model the interactions of human operator decisions, TBM behaviors, and ground conditions in an integrated representation. The study also shows that the multivariate Random Forest (MRF) algorithm can concurrently make multiple decisions on TBM steering control parameters during driving.

This dissertation has demonstrated that TBM data contain valuable information that can be extracted to benefit tunneling operations. Due to the complex relationships within the data, the nonlinear and nonparametric machine learning models offer advantages over the conventional linear and parametric models. This development is envisioned to be a building block that advances the development of autonomous TBM technology.

Understanding of fluid behavior in fluid-saturated porous media has been a challenging research problem tackled through theoretical, experimental, and numerical approaches over many decades. Modeling such multiphysics phenomena, in both small and large deformation regimes, requires careful consideration of numerical accuracy and stability to achieve reliable results. This study employs the Material Point Method (MPM), a hybrid Eulerian-Lagrangian approach that can simulate large deformation problems involving history-dependent materials. The dissertation's contributions can be divided into two main application areas: (i) deep geomechanical well analysis under fluid injection and extraction, and (ii) surficial geotechnical structures involving the nonlinear interaction of free-surface flow and deformable porous media.

First, a new approach called CutMesh MPM is introduced for small-deformation solid mechanics. This method improves standard MPM discretization by re-evaluating quadrature weights and incorporating explicit boundary discretization to better track evolving surface geometry and apply boundary conditions. Algorithms for boundary detachment, detection, and reconstruction are introduced to track the evolution of wellbore geometry due to material detachment. The efficacy of the approach is validated through numerical studies of dry boreholes under anisotropic loading. Building on this, the study extends the CutMesh MPM framework to hydro-mechanical analysis of saturated boreholes. The coupled formulation integrates a two-phase single-point MPM formulation, assuming a fractional-step method, for accurate analysis of fluid extraction and injection in poroelastic and poro-elasto-plastic materials.

The second part of the dissertation focuses on modeling free-surface fluid dynamics, seepage, and interactions with porous solids. A mixed MPM formulation is introduced to address challenges related to the numerical instability of incompressible fluid flow. Several stabilization strategies are proposed and utilized to enhance the simulation's robustness and accuracy. This formulation is verified and validated through extensive numerical tests on complex fluid dynamics problems. Additionally, a unified approach integrating the Navier-Stokes and Darcy-Brinkman-Forchheimer equations is proposed to model flows in both non-porous and porous domains. This method uses the blurred interface technique to improve numerical stability at the interface, as demonstrated in a range of 1D, 2D, and 3D benchmarks. Finally, the mixed MPM is extended to account for strongly coupled hydro-mechanical problems through a three-field formulation involving solid displacement, fluid displacement, and pressure. The proposed approach features stabilized discretization methods and consistent derivations, with preliminary results in three-dimensional consolidation problems demonstrating its potential.

The geotechnical engineering field is full of large deformation problems, ranging from settlements caused by human constructions to naturally occurring landslides. Modeling these problems is therefore of paramount importance for risk management and improved understanding of soil/rock behavior. Many particle-based or mesh-free methods have been developed to reliably and efficiently analyze large deformation problems. One such method is the Material Point Method (MPM) (Sulsky et al., 1994; 1995), which is a hybrid Eulerian-Lagrangian approach capable of simulating large deformation problems of history-dependent materials. While the MPM can represent complex and evolving material domains by using the Lagrangian material points, boundary conditions are often applied to the Eulerian nodes of the background mesh. Consequently, the use of a structured mesh may become prohibitively restrictive for modeling complex boundaries such as a landslide topography.

This research explores the suitability of unstructured background mesh with isoparametric elements to model irregular boundaries in the MPM. An inverse mapping algorithm is used to transform the material points from the global coordinates to the local natural coordinates. Dirichlet velocity and frictional boundary conditions are applied in the local coordinate system at each boundary node. This approach of modeling complex boundary conditions is validated by a patch test, modeling of a gravity-driven rigid block sliding on an inclined plane, and performing mesh convergence study on the stress obtained in a hollow cylinder problem with pressure applied on the outer boundary.

The proposed approach is then applied to a sand production problem. 2D MPM with isoparametric elements is used to simulate two different sets of laboratory experiments labeled the PEA-135 and PTC which investigate sand production phenomenon on Castelgate Sandstone. The material points representing the material volume (or mass) are allowed to be removed, representing the disintegration of sandstone rock to sand. This removal is dictated by a criterion developed based on the equivalent plastic deviatoric strain accumulated. To the author's best knowledge, this is the first sand production study using the MPM. A new constitutive model, the Bonded Norsand, is developed to accurately describe the behavior of the sandstone material. Calibration of the parameters are done through available laboratory triaxial drained compression tests on the material.

This method is later applied to a flume test of controlled dry granular flow on an inclined plane conducted by the United States Geological Survey (USGS). This flow problem provides an opportunity to demonstrate the ability of the proposed approach to model large deformation problems with high velocity or kinetic energy in contrast with the sand production problem. Mesh sensitivity and time step effects are investigated. The approach is able to capture most of the mechanisms of dry granular flow. Comparisons are also made to other approaches within the MPM, as well as other continuum-based methods such as the Finite Element Method (FEM) and Smooth Particle Hydrodynaics (SPH).

Lastly, an expansion of this approach to 3D is presented where the inverse-mapping is solved iteratively and demonstrated with three examples. The first is the 3D version of the previously discussed flume test of controlled dry granular flow on an inclined plane. The second is an experiment performed on the same flume with the same dry sand but with a slit gate to demonstrate the ability to model variations in 3D. Lastly, the approach is used to model a historic debris flow, the Yu Tong Road landslide in Hong Kong. The approach is found to be able to simulate the mechanisms of these different cases. Comparisons with the results from other continuum-based methods such as the Arbitrary Lagrangian-Eulerian approach and the SPH are also made in this section.

Geotechnical engineering, more so than any other branch of engineering, relies on measurement for the design and execution of our works. The natural materials that form the basis of geotechnical engineering design cannot be specified in a factory – they were formed through geologic processes and are already in place, awaiting discovery by the engineer. To characterize their properties, the materials must be measured. The execution and performance of the design must also be measured if it is to be confirmed to conform to the quality and behavior expected by the engineering team. Since the subsurface structures cannot be directly observed, we rely on instrumentation to measure their behavior.Distributed Fiber Optic Sensing (DFOS) refers to a range of technologies that enable the measurement of physical phenomenon along a fiber optic cable. DFOS relies upon the fiber optic cable as both the sensor as well as the data conduit. By carefully interrogating changes in light that is sent through the fiber optic cable, measurements such as strain and temperature can be taken in a continuous readout along the cable. This technology has been in development for several decades and has been adopted for applications within civil engineering, including geotechnical engineering. The unique ability to generate continuous measurement profiles offers a significant benefit over conventional point-based instrumentation, particularly in applications where the strain and temperature fields are changing in non-linear and unpredictable ways. This research explores the use of DFOS for the monitoring of strain and temperature in deep foundations during loading and vertical soil displacements. These applications are particularly suited to the benefits that DFOS offers, specifically the continuous strain readout that is produced by fiber optic monitoring. Best practices for installation and monitoring were developed, and these advancements were applied on a series of field projects to ascertain their performance and gain insight into the behavior of the monitored assets. The applications included monitoring of strain development in deep foundations during post grouting, monitoring of strain in deep foundations during top-down load testing, monitoring of blanket uplift on a river levee, and monitoring of vertical ground deformations during surcharge loading of a soft clay site. In each case, the author developed novel improvements for the design, installation, monitoring, and interpretation of DFOS data to provide new insight into the geomaterial and geostructure behavior. In the application of strain measurements in deep foundations, three major improvements and new findings were explored and verified. For the installation phase, the current state of practice of pre-tensioning strain cables prior to concreting was replaced with installation with slack removed but without any additional pre-tension. The encapsulation in concrete was proved to be enough to fully transfer compressive strains within the foundation element to the fiber optic sensing cable through validation data using conventional point-based strain measurement. During the data processing, installation of 4 vertical strain cables over the standard 1 or 2 verticals in the literature allowed for identification and analysis of non-uniform strain distribution across the pile at single depths. This variation has often been ignored or discarded in favor of averaging the strain measurements across the pile, missing real non-idealized pile behavior during loading, or incorrectly discarding valid data. Finally, lab and field testing was performed to capture and quantify the effect of fiber optic analyzer architecture on the processed strain readout when the strain along the sensing fibers is not constant during the reading interval. This effect has not been identified or explored and was shown to have a quantifiable and correctable impact on the data, as well as highlighting an additional performance metric to consider when selecting a fiber optic analyzer for a particular application. In the application of DFOS for soil vertical strain monitoring, two improvements on the state of practice were developed. The first is the detailed exploration and demonstration of the benefit of continuous strain measurements over point-based measurements in the identification and quantification of local subsurface strain. The highly local nature of these strains, as well as the judgement required in the placement of point-based measurement points, result in DFOS offering a clear advantage over the standard state of practice. The necessity for temperature compensation of subsurface DFOS strain measurements was also explored and quantified. The current state of practice is to treat the subsurface temperature as stable, however measurements show that surface temperature fluctuations can impact the soil temperature to significant depths and must be corrected prior to integration of the distributed strain profiles to calculate cumulative surface displacement.

The micro-scale mechanical behavior of civil infrastructure and the earth immediately sur-rounding it is indicative of material properties, condition and use level. In addition, the geometry of infrastructure systems can be complex and spanning large extents. Phase- sensitive optical time domain reflectometry (φ-OTDR), also known as distributed acoustic sensing (DAS) or distributed vibration sensing (DVS), is a revolutionary technology that uses interferometry within optical fibers to make distributed measurements of strain at higher precision than ever before possible. This technology can measure extremely small strains (nanostrain level) at high spatial resolutions (several meters) and over long distances (10’s of kilometers). Since φ-OTDR can measure such small perturbations, it has gained widespread attention within the earth and planetary science community for measuring seismic wave propagation. These measurements have been predominantly focused on capturing the phase behavior of seismic waves at a crustal or exploration scale. However, for application to mea- suring dynamic strain in civil infrastructure the measurements must be understood for their strain magnitude as well as phase behavior. This dissertation serves to demonstrate the use of φ-OTDR for monitoring civil infrastructure systems through the presentation of several practical verification cases. Optical fibers were coupled to a soil site, a pavement system, and a superstructure to demonstrate how φ-OTDR can be used to quantitatively measure dynamic strain in these elements. The φ-OTDR measurements were verified through the use of velocity transducers (geophones), strain gauges, and another distributed fiber optic sens- ing technique called optical frequency-domain reflectometry (OFDR). Numerical modeling of φ-OTDR measurements is demonstrated for describing the measurements in a pseudo- static case as well as during stress-wave propagation. It is shown that φ-OTDR provides capabilities for monitoring civil infrastructure that currently-used techniques cannot due to its superior measurement precision and high acquisition rates. φ-OTDR is demonstrated as a new tool for monitoring civil infrastructure.