INTEGRATED DYNAMIC MANAGEMENT OF METHANE EMISSIONS IN INDUSTRIAL ENERGY SYSTEMS: A PHYSICS-INFORMED TRANSFORMER DIGITAL TWIN FRAMEWORK FOR REAL-TIME VIRTUAL SENSING AND CONTROL OF METHANE

Abstract

Reciprocating engines powered by natural gas (NGFREs) play a central role in the natural gas transmission network. They serve as a significant source of methane slip and constitute an ongoing reliability bottleneck. Given methane’s immediate warming potential, increasing regulatory restrictions, and the extensive fleet of existing engines, it is now imperative to approach emissions monitoring and maintenance as integrated rather than isolated activities. The fundamental premise of this work is that emissions management and reliability enhancement should be addressed concurrently. Nevertheless, most compressor stations continue to operate with limited sensing capabilities, legacy control systems, constrained on-site computational resources, and minimal labeled failure data. These limitations hinder the deployment of conventional emissions analysis tools and data-intensive predictive maintenance solutions that are affordable, transparent to operators, and effective across various engine configurations.This dissertation develops an integrated dynamic management framework for methane emissions and equipment health in industrial NGFRE systems. The work begins with a cross-disciplinary review of methane measurement technologies and NGFRE predictive maintenance methods, showing that existing solutions either depend on expensive, maintenance-intensive analyzers or focus narrowly on fault detection without treating emissions as a first-class performance variable. Building on these observations, the dissertation introduces the Integrated Dynamic Management of Complex Systems (IDMCS) framework for NGFRE: a design-science architecture that links experimental engine campaigns, cyber-secure data acquisition, physics-guided machine learning, and edge–cloud digital twins into a single methodology for observing, predicting, and managing NGFRE complex behavior. Within this framework, the first significant technical contribution is a unified, stroke-agnostic methane virtual sensor. Using data from both large-bore two-stroke and four-stroke NGFREs, the dissertation reconstructs and extends the earlier transformer-based methane model and embeds explicit combustion and thermodynamic constraints through physics-informed loss functions. To overcome generalization limits and provide short-horizon forecasts, the work further integrates a time-series foundation model that predicts methane slip several minutes ahead using standard supervisory control and data acquisition (SCADA) signals. The resulting unified methane virtual sensor is designed for real-time and continuous deployment and emphasizes statistical accuracy, physical plausibility, and cross-engine transferability. xix The second significant contribution is a predictive maintenance and emissions-aware digital twin that operates under realistic industrial constraints. Using structured-text logic on an industrial programmable logic controller, the dissertation implements an end-to-end pipeline that streams real-time results of various parameters and presents interpretable dashboards and alerts for operators. Validated across laboratory testbeds and two commercial compressor stations, this architecture ingests heterogeneous data, tolerates communication and power limitations, and surfaces actionable information in time for maintenance crews to respond. The third contribution is a closed-loop air–fuel ratio control system that algorithmically adjusts the engine’s bypass valve using a linear regression machine learning model for engine load estimation, running on an industrial PLC. The IDMCS framework, the unified methane virtual sensor, and the edge–cloud digital twin together offer a coherent, field-tested path from retrospective emissions reporting and reactive maintenance toward real-time virtual sensing and proactive intervention in industrial natural gas energy systems. The research also yields a patent-pending methane virtual sensor artefact and a real-time closed-loop optimization artefact under preparation for invention disclosure. It outlines a roadmap of additional digital tools, including unified VOC sensing, automated EPA reporting workflows, and NOₓ/O₂ sensor remaining useful life estimation, that extend the same architecture beyond methane-focused monitoring and control.