Recurrent pipeline failures continue to be a source of safety and economic risk related to processing, transporting, and distributing natural gas. Studies have shown the lack of comprehensive, integrated, and accessible risk-informed integrity management models and tools for pipeline operators is a major contributor. To address this gap, this research presents a system-level Prognosis and Health Monitoring (PHM) modeling framework for gas pipeline system integrity management to prevent or reduce the likelihood of failures. The proposed PHM approach takes into consideration all possible failure modes of the pipeline under study. It leverages the advancement of sensor technology to stream field data in real-time to perform a dynamic system-level failure analysis based on Hybrid Causal Logic (HCL) including a Dynamic Bayesian Network (DBN) corrosion model, to provide cost-effective and optimal mitigation actions such as sensor placement and maintenance schedule optimizations. The developed models are implemented in a software platform where the pipeline operators can observe the real-time and projected health state of the pipeline and the set of suggested actions to enhance the structural integrity of the pipeline system. The platform includes three main modules: Real-Time Health Monitoring, System-Level Reliability, and Optimal Mitigation Actions. From a safety perspective, the proposed PHM can prevent pipeline failures or reduces their likelihood by supporting pipeline operators in optimal decision-making and planning activities. To demonstrate potential benefits and performance of the proposed framework and software implementation, it is applied in a case study involving a corroding gas transmission pipeline.

Resistive random access memory (RRAM) is one of the emerging technologies of non-volatile memory that attracts excellent research interest. This research uses a Kinetic Monte Carlo (KMC) simulation to demonstrate the switching physics in the ???x-based Ox-RRAM, including forming, SET, and RESET processes. A dynamic resistor network has been built to simulate oxygen vacancy generation, oxygen vacancy, oxygen ion recombination, oxygen ion migration, and hopping during resistive switching. The conductive filament (CF) configuration resulting from oxygen vacancy distribution is updated in time steps. The proposed KMC model can provide insight into RRAM electronic characteristics based on its microstructural material behaviors. The switching layer was divided into resistors with different resistances. The research aims to use this model as the foundation to predict the performance and reliability of RRAM devices, which covers performance fluctuation during switching, resistance state degradation (endurance), cycle-to-cycle, device-to-device variation, retention, and window margin trade-off. This simulation model provides a tool for comparing parameter significance and predicting device behaviors. This research also explores implementing the Bayesian approach that can integrate heterogeneous information sources and modify the simulation results.

Dynamic Probabilistic Risk Assessment (PRA) refers to an emerging class of PRA methods that generate risk scenarios through the model-based simulation of systems such as nuclear power plants (NPPs) and their crew response to accident initiators. The dynamic PRA approach offers several advantages over the conventional approaches currently used by the nuclear industry worldwide. These advantages include: (1) time-dependent prediction of the operator error-forcing contexts, (2) better representation of the thermal-hydraulic success criteria, and (3) considerable reduction in analyst-to-analyst variability of the results. An example of such a simulation platform is the Accident Dynamics Simulator coupled with the Information, Decision and Action in a Crew context cognitive model (ADS-IDAC), and a realistic NPP thermal-hydraulic model. The aim of this research is to integrate qualitative and quantitative hybrid causal logic into the ADS-IDAC dynamic PRA platform. This makes ADS-IDAC a more practical and realistic analysis tool for specific applications. These applications are primarily event assessments, but also include the ability to analyze highly dynamic and complex accident scenarios in support of conventional PRAs. This work offers major modeling enhancements of ADS-IDAC, including dynamically linked fault trees (FTs) for support and frontline systems modeling, more advanced system and operating crew modeling capabilities, comprehensive quantification features modeling human failure evens (HFEs), and uncertainty propagation through the generated discrete dynamic event tree (DDET). The new risk assessment process was streamlined with the help of a newly developed user-friendly graphical interface, which provides efficient and convenient access to all the capabilities of the ADS-IDAC simulation engine.

Pipeline integrity management refers to an approach of understanding and operating pipelines in a safe and reliable manner. In this work, firstly, a probabilistic predictive model for internal corrosion of natural gas pipelines subject to aqueous CO2/H2S environment has been proposed. The model regards uniform and pitting corrosion as two main corrosion mechanisms and has been calibrated with the experimental data in a deterministic framework. Methodologies of simulating and accounting for temporal and spatial variabilities of operating parameters have been proposed and applied to the model for field applications. The model has been validated against field data from eight wet gas gathering pipelines in a probabilistic framework. Secondly, a reinforcement learning (RL)-based maintenance scheduler has been proposed for pipeline maintenance optimization problems by leveraging the proposed predictive corrosion model and the Q-learning and Sarsa algorithms. A case study has shown the superiority of the proposed maintenance scheduler over the periodic maintenance policy in reducing the maintenance costs. Finally, the previous two parts of work have been integrated into a pipeline system integrity management software featuring pipeline health monitoring, corrosion prognosis, system-level failure analysis, sensor placement optimization, and inspection/maintenance optimization. A case study has been provided to demonstrate the capabilities of the software.

We live in the era of big data in which the advancement of sensor and monitoring technologies, data storage and management, and computer processing power enable us to acquire, store and process over 2.5 Quintilian bytes of data daily. This massive data brings the necessity of using trustable and high-performance data-driven models that extract knowledge out of data. This dissertation focuses on learning to solve highly risk-averse and complex sequential decision-making problems from retrospective data sets by deep Reinforcement Learning (RL).

Deep RL has gained remarkable breakthroughs in many applications. It achieved superhuman performance in video and Atari games, defeated the world champion in game of Go, gained competent autonomy in simulated self-driving cars, and successfully learned to perform some robotic tasks. Despite all the notable advancements in deep RL, its application to real-world problems such as clinical treatment policy or industrial asset maintenance management is insignificant. Studies are underway to investigate deep RL use in realistic problems; however, none has been deployed in real-world settings. Several limitations hinder the deep RL application to real-world problems, among which trustability and excessive thirst for data are the main issues. This research is an effort to smooth the way of applying deep RL to real-world problems by addressing the above two limitations.

We first provide a concrete definition for trust in RL algorithms, and then we propose a sample efficient deep RL agent that computes a trustable solution to real-world sequential decision-making problems. The agent tackles the trust problem from two aspects. It imposes risk barriers to the RL agent's policy improvement process and provides off-policy performance estimation with a confidence bound prior to putting the agent in interaction with the actual system or environment. We address the RL significant demand for data by implementing the most advanced efficient data utilization techniques as well as deploying new techniques that improve the Trustable Deep RL sample efficiency.

The proposed methodology is tested and evaluated on a novel pipeline corrosion maintenance test bench that mimics the real system restrictions. The results witness that the Trustable Deep RL algorithm efficiently digests a retrospective data set from the pipeline environment and gains a superior and trustable interaction policy.

The use of composite pipelines has grown in recent years as an alternative to steel in the oil and gas industry, due to their excellent corrosion resistance and high specific strength. Among these developments, thermoplastic composite pipes have seen a recent rise in research and use due to their greater flexibility and impact resistance than traditional thermoset-matrix composites. To understand and predict the performance of these materials in the field, studies have derived experimental, analytical, and theoretical estimates of the mechanical and thermal properties of thermoplastic composite pipes (TCP) and reinforced thermoplastic pipe (RTP) in dry conditions, but studies in wet or corrosive environments are currently restricted to the empirical domain. A better understanding of the performance of thermoplastic-matrix composites in marine and water-saturated environments is needed, because 30% of the world’s current oil and gas supply is produced offshore; the offshore portion of the upstream segment of the oil and gas industry has risen dramatically in recent years, and this trend is expected to continue [1], [2].To address part of this gap in modeling capability, this research used a probabilistic approach to predict degradation rates of the constituent polymer matrix and reinforcement fibers within an individual composite ply given the local temperature and the amounts of absorbed fluids. This method was applied by developing a TCP simulation code, which consists of three major time-dependent processes: 1) absorption of oil- and water-based fluids for each layer of the pipe given spatial and temporal variations in the diffusivity parameters, 2) evaluation of the local composite properties given the amount of degradation to the matrix and fibers during each time window, and 3) prediction of the stresses at each ply under defined thermomechanical loading conditions given the current state of each ply’s mechanical properties and subsequent determination of ply failures. The ply failures determined from the third process incur further reduction in the stiffness of the failed ply, resulting in greater amounts of stress applied to the remaining un-failed plies until all plies in the laminate have failed, thus constituting a progressive failure mechanism. In order to model these processes while accounting for various uncertainties in material and environmental inputs, new methods for estimating certain process factors, such as diffusivity and the extent of composite stiffness and strength reduction following degradation and ply failure, are developed using comparisons to real-world data. Similarly, to verify the validity of each of the three major physical processes in this simulation, the results of each given set of controlled inputs are compared to experimental data on TCP and RTP from various lab and field experiments. It was found that in cases where the observed degradation was low, agreement with the model was generally good, but the model tended to underestimate the amount of degradation in the more severe cases. Future work will aim to improve the prediction capabilities of this method by adjusting the existing processes in the model and incorporating factors that the model currently does not consider, such as fillers and the pH of absorbed fluids. Finally, the model presented herein has been implemented as a new module in a pre-existing online application initially developed to predict the reliability of metallic pipes in a real world environment, in order to provide insights to the oil and gas industry on the risks of certain locations. The current implementation for the non-metallic pipes includes a prediction of the failure rates, mechanical stiffness degradation, and fluid absorption for segments of pipe, with plans to add maintenance and sensor placement features in the future.

Degradation of cable insulation subject to heat and radiation is modeled by physics-based chemical, mechanical, and electrical approaches. Experimental data of cross-linked polyethylene (XLPE), ethylene propylene rubber (EPR), and silicone rubber (SIR) are incorporated to validate the models.

Chemical approaches divide aging kinetics into reaction- and diffusion-controlled mechanisms rendering homogeneous and heterogeneous cross-sections, respectively. The degradation profile of a cross-section serves as the base of Dichotomy Model. The model dichotomizes one bulk material into virtually degraded and non-degraded parts. The quotient of the two parts is determined by a diffusion theory in diffusion-controlled cases or by an exponential distribution in reaction-controlled scenarios to predict the lifespan of the cable insulation. Besides the degradation of the insulation, the migration and decomposition of the antioxidant in the insulation are also modeled by a reaction-diffusion theory representing the uphill diffusion of the antioxidant. The uphill diffusion is driven by the unevenly-distributed activity coefficient linearly proportional to oxygen concentration with a negative slope.

Both mechanical and electrical approaches are derived from Dichotomy Model. The degradation of mechanical performance is focused on elongation at break (EAB) while that of electrical property is quantified by resistance. EAB and resistance as functions of time are deterministically developed. Incubation time is a parameter of the function representing the time period when the changes of the physical properties are negligible. Drop-off rate is the other parameter describing the trends of the properties. The values of the two parameters follow the trends of temperature and dose rates, which makes the functions predictable in different aging conditions. Unlike the state of the art, this model can accommodate the shape change of the curves of EAB and resistance along a time axis.

By Bayesian parameter estimation, the deterministic function has been converted into a probabilistic equation. This method can represent the thresholds of the EAB and resistance in field application with uncertainty denoting the reliability of cable insulation by expected lifespan in the form of probability. The approach is the first in the area.

Water distribution systems are vital to the well-being of communities because they contribute to the functionality of all other infrastructure and lifeline systems. Earthquakes and other natural hazards can cause damage to the components of a water distribution system, causing far-reaching socioeconomic consequences. This research begins with the development of an end-to-end simulation framework to model post-earthquake functional loss and restoration of a water system, which encompasses seismic hazard characterization, component damage assessment, hydraulic performance evaluation, and network restoration modeling. The modeling framework is validated using data from the 2014 South Napa Earthquake and extended to a hypothetical scenario. The end-to-end simulation framework is then extended to consider stochastic event set assessments of the water network using the UCERF2 (Uniform California Earthquake Rupture Forecast, Version 2) earthquake rupture forecast model. Given that the end-to-end performance evaluation of distributed infrastructure for a large set of events is computationally expensive, a framework that uses Active learning to select a subset of ground motion maps and associated occurrence rates that reasonably estimates the water network risk is also developed. To deal with the temporal complexities that are embedded in the post-earthquake restoration process, a dynamic updating methodology is developed to reduce uncertainties in the outcomes of post-event recovery forecasts using Bayesian Inferencing, by exploiting real-time data. The specific example of updating predictions (post-earthquake functional recovery forecasts including total recovery time and complete recovery trajectory) is presented and validated on a real pipe network (Napa water system) and event (2014 earthquake and recovery). Ultimately, the frameworks and models developed as part of this work can inform risk-based decision making and resilience planning of water networks and other lifeline systems.

As we shift away from fossil fuel-derived energy and the electrification of various systems grows exponentially, society is increasingly relying on energy storage systems such as batteries to store and distribute energy on demand. While batteries are a widely studied technology, other energy storage technologies such as supercapacitors have unique performance qualities that can provide alternatives or supplements to batteries in systems. This dissertation delves into MnO2 pseudocapacitors, aiming to unravel the degradation mechanisms impacting their performance and reliability under diverse operating conditions, toward their usage (or other pseudocapacitive materials) in real systems. While there is a growing body of research on pseudocapacitors, including MnO2-based pseudocapacitors, little work has gone into frameworks for predicting their behavior under various environmental and operating conditions. Through a multifaceted approach of experimental investigation, physics-based modeling, and probabilistic analyses, this research centers on elucidating the impact of temperature and degradation of active material on the aging of devices, ultimately predicting capacitance reduction over time.

Two approaches are taken in this research to predict end of life for pseudocapacitors. The first approach utilizes a first-principles physics model, augmented with Bayesian methods to integrate experimental and statistical information with uncertainties into the model. This method allows for the prediction of changes happening in the cell over numerous cycles, and the associated capacitance reductions. The second approach develops for the first time the use of a Bayesian Monte Carlo approach for estimating the remaining useful life (RUL) of pseudocapacitors based on experimental data with quantified uncertainty. The combination of estimates from these models can lead to improved prediction based both on real world experience through data and on theoretical knowledge of the system.

Overall, this dissertation advances the prediction of aging in MnO2 pseudocapacitors and develops models that can be applied to various other pseudocapacitive systems. These contributions lay a foundation for further exploration that may decrease the time required to conduct pseudocapacitor experiments and contribute to the management and longevity of energy storage systems containing pseudocapacitors.

Incorporating automated driving technologies in commercial heavy-duty operations aims to increase traffic efficiency and safety. However, developers, fleet operators, and regulators must address the unique safety risks that Automated Driving System (ADS) technology introduce. Potential Heavy-Duty Automated Vehicle (HD-AV) fleet operations envision a team of human and machine agents, including the ADS, an onboard safety driver, a fleet operations center, and, in some cases, an onboard safety operator. The complex interactions between these human and machine agents must be addressed when determining the system’s safety requirements and design. Safety metrics usually focus on ADS performance, but to adequately inform system design and safety requirements, these metrics must also focus on human-system interactions. This work proposes a system model-based approach to develop human-system interaction metrics focusing on hardware, software, and human impacts on the operational safety of automated driving system fleets.