{"id":{"repo_id":"houston","oai_identifier":"oai:uh-ir.tdl.org:10657/20708"},"canonical_url":"https://search.dev.ndltd.org/etd/houston/oai:uh-ir.tdl.org:10657/20708","repository":{"repo_id":"houston","name":"University of Houston","base_url":"https://uh-ir.tdl.org/server/oai/request"},"display":{"title":"Machine Learning Enhanced Power Converter Design and Prognostic Health Monitoring in DC Distribution Systems","abstract":"Power electronics has become an indispensable part of any modern electrical system, ranging from small to large-scale applications requiring efficient energy conversion at every power distribution stage. However, new challenges emerge as the converter-based systems grow in complexity, i.e., from individual power converter design and control to large-scale management of a series of interconnected converters in DC microgrids and DC distribution systems such as data centers, power grids, etc.. Though mature, well-established, and widely adopted, traditional model-based methods often struggle to address real-world challenges such as nonlinear system dynamics, uncertainties in the system parameters, and the growing complexity of interconnected converters in small to large-scale DC distribution systems. Moreover, real-time anomaly detection, health monitoring, and operational conditioning in such evolving systems require fast, adaptive, and scalable solutions to handle diverse operating conditions and unpredictable disturbances. Machine Learning (ML) provides a transformative toolset to tackle such inevitable real-world challenges prevalent in different power electronics domains. While traditional model-based approaches rely entirely on several stated assumptions and parameter estimations, data-driven models such as ML can capture complex system behaviors without solely relying on the idealized physics-based model, enabling predictive control, fault detection, and optimization under uncertainty scenarios. In converter design, ML can accelerate topology selection and parameter tuning; in operational control, it can facilitate adaptive decision-making and multi-objective optimization; and in large-scale DC systems, it can enable coordinated energy management across multiple nodes with varying priorities and constraints. ML can process high-frequency measurement data in real time for anomaly detection and health monitoring to identify subtle deviations indicative of incipient faults—capabilities that exceed the responsiveness of many traditional detection schemes. Compared to conventional techniques, machine learning approaches inherently adapt to changing system dynamics, scale efficiently to high-dimensional problems, and can integrate diverse sources of information—from sensor streams to historical operational data—into unified decision frameworks. These advantages make ML particularly well-suited for addressing the evolving demands of power electronic systems where flexibility, predictive capability, and operational resilience are crucial. This dissertation advances machine learning applications in power electronics by proposing frameworks for optimal control, real-time in-situ health monitoring, predictive energy management, and hardware-accelerated on-edge inference in converter-dominated DC distribution systems. Validated through hardware-in-the-loop (HIL) and FPGA implementations, the methods deliver significant improvements in adaptability, computational efficiency, and fault resilience over traditional rule-based approaches. Chapter 1 addresses the lack of a unified view on ML's role in power electronics by performing a 2000-2024 bibliometric and thematic analysis, identifying the shift from heuristic and classical ML dominance to emerging methods like Graph Neural Networks (GNNs) and Generative Models (GenAI), and pinpointing open research gaps. Chapter 2 tackles the slow, manual converter design workflow by introducing a computer vision and OCR-based digitize-and-simulate pipeline that converts electrical schematics into a netlist for automated simulation, cutting design time from weeks to seconds. Chapter 3 responds to the challenge of limited topology innovation by building a structured graph-based dataset and applying Graph Variational Autoencoders and fine-tuned GPT-2 models to generate novel, high-performance converter designs. Chapter 4 addresses the difficulty of scalable, real-time grid health assessment by proposing an FPGA-accelerated, impedance-based monitoring method using neural network regression and classification to detect loading conditions and faults without distributed sensors or communication latency. Chapter 5 targets suboptimal energy scheduling in variable conditions through optimization-based techniques, i.e., mixed integer linear programming (MILP), and reinforcement learning-based techniques, i.e., Deep Q-Networks (DQN) agents that leverage intermittent renewable sources such as photovoltaic (PV), dynamic load profiles, and volatile electricity prices. Together, these contributions advance the state of the art from manual, expert domain knowledge-centric workflows toward fully autonomous DC distribution systems whose converters can be generated, verified, monitored, and dispatched by learning algorithm—paving the way for more resilient power converter-dominated DC distribution systems.","abstract_html":"Power electronics has become an indispensable part of any modern electrical system, ranging from small to large-scale applications requiring efficient energy conversion at every power distribution stage. However, new challenges emerge as the converter-based systems grow in complexity, i.e., from individual power converter design and control to large-scale management of a series of interconnected converters in DC microgrids and DC distribution systems such as data centers, power grids, etc.. Though mature, well-established, and widely adopted, traditional model-based methods often struggle to address real-world challenges such as nonlinear system dynamics, uncertainties in the system parameters, and the growing complexity of interconnected converters in small to large-scale DC distribution systems. Moreover, real-time anomaly detection, health monitoring, and operational conditioning in such evolving systems require fast, adaptive, and scalable solutions to handle diverse operating conditions and unpredictable disturbances. Machine Learning (ML) provides a transformative toolset to tackle such inevitable real-world challenges prevalent in different power electronics domains. While traditional model-based approaches rely entirely on several stated assumptions and parameter estimations, data-driven models such as ML can capture complex system behaviors without solely relying on the idealized physics-based model, enabling predictive control, fault detection, and optimization under uncertainty scenarios. In converter design, ML can accelerate topology selection and parameter tuning; in operational control, it can facilitate adaptive decision-making and multi-objective optimization; and in large-scale DC systems, it can enable coordinated energy management across multiple nodes with varying priorities and constraints. ML can process high-frequency measurement data in real time for anomaly detection and health monitoring to identify subtle deviations indicative of incipient faults—capabilities that exceed the responsiveness of many traditional detection schemes. Compared to conventional techniques, machine learning approaches inherently adapt to changing system dynamics, scale efficiently to high-dimensional problems, and can integrate diverse sources of information—from sensor streams to historical operational data—into unified decision frameworks. These advantages make ML particularly well-suited for addressing the evolving demands of power electronic systems where flexibility, predictive capability, and operational resilience are crucial. This dissertation advances machine learning applications in power electronics by proposing frameworks for optimal control, real-time in-situ health monitoring, predictive energy management, and hardware-accelerated on-edge inference in converter-dominated DC distribution systems. Validated through hardware-in-the-loop (HIL) and FPGA implementations, the methods deliver significant improvements in adaptability, computational efficiency, and fault resilience over traditional rule-based approaches. Chapter 1 addresses the lack of a unified view on ML's role in power electronics by performing a 2000-2024 bibliometric and thematic analysis, identifying the shift from heuristic and classical ML dominance to emerging methods like Graph Neural Networks (GNNs) and Generative Models (GenAI), and pinpointing open research gaps. Chapter 2 tackles the slow, manual converter design workflow by introducing a computer vision and OCR-based digitize-and-simulate pipeline that converts electrical schematics into a netlist for automated simulation, cutting design time from weeks to seconds. Chapter 3 responds to the challenge of limited topology innovation by building a structured graph-based dataset and applying Graph Variational Autoencoders and fine-tuned GPT-2 models to generate novel, high-performance converter designs. Chapter 4 addresses the difficulty of scalable, real-time grid health assessment by proposing an FPGA-accelerated, impedance-based monitoring method using neural network regression and classification to detect loading conditions and faults without distributed sensors or communication latency. Chapter 5 targets suboptimal energy scheduling in variable conditions through optimization-based techniques, i.e., mixed integer linear programming (MILP), and reinforcement learning-based techniques, i.e., Deep Q-Networks (DQN) agents that leverage intermittent renewable sources such as photovoltaic (PV), dynamic load profiles, and volatile electricity prices. Together, these contributions advance the state of the art from manual, expert domain knowledge-centric workflows toward fully autonomous DC distribution systems whose converters can be generated, verified, monitored, and dispatched by learning algorithm—paving the way for more resilient power converter-dominated DC distribution systems.","abstract_has_math":false,"creators":["Bohara, Bharat 1991-"],"institution":"University of Houston","degree_name":"Doctor of Philosophy","degree_level":null,"degree_discipline":"Electrical Engineering","degree_department":null,"school":null,"contributors":[],"advisors":["Krishnamoorthy, Harish Sarma"],"committee_chairs":[],"committee_members":["Nguyen, Hien V.","Fan, Lei","Shi, Jian","Rajashekara, Kaushik"],"year":2025,"date_issued":"2025-08","date_published":"2025-08","updated_at":"2026-07-24T02:32:32Z","subjects":["Electrical engineering"],"languages":["English"],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://hdl.handle.net/10657/20708","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Krishnamoorthy, Harish Sarma"]},{"key":"dc:contributor.committeemember","label":"Committee Member","values":["Nguyen, Hien V.","Fan, Lei","Shi, Jian","Rajashekara, Kaushik"]},{"key":"dc:creator","label":"Author","values":["Bohara, Bharat 1991-"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2025-10-06T20:00:42Z"]},{"key":"dc:date.issued","label":"Date","values":["2025-08"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Electrical Engineering"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Doctor of Philosophy"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["University of Houston"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Electrical engineering"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["English"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/10657/20708"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Power electronics has become an indispensable part of any modern electrical system, ranging from small to large-scale applications requiring efficient energy conversion at every power distribution stage. However, new challenges emerge as the converter-based systems grow in complexity, i.e., from individual power converter design and control to large-scale management of a series of interconnected converters in DC microgrids and DC distribution systems such as data centers, power grids, etc.. Though mature, well-established, and widely adopted, traditional model-based methods often struggle to address real-world challenges such as nonlinear system dynamics, uncertainties in the system parameters, and the growing complexity of interconnected converters in small to large-scale DC distribution systems. Moreover, real-time anomaly detection, health monitoring, and operational conditioning in such evolving systems require fast, adaptive, and scalable solutions to handle diverse operating conditions and unpredictable disturbances. Machine Learning (ML) provides a transformative toolset to tackle such inevitable real-world challenges prevalent in different power electronics domains. While traditional model-based approaches rely entirely on several stated assumptions and parameter estimations, data-driven models such as ML can capture complex system behaviors without solely relying on the idealized physics-based model, enabling predictive control, fault detection, and optimization under uncertainty scenarios. In converter design, ML can accelerate topology selection and parameter tuning; in operational control, it can facilitate adaptive decision-making and multi-objective optimization; and in large-scale DC systems, it can enable coordinated energy management across multiple nodes with varying priorities and constraints. ML can process high-frequency measurement data in real time for anomaly detection and health monitoring to identify subtle deviations indicative of incipient faults—capabilities that exceed the responsiveness of many traditional detection schemes. Compared to conventional techniques, machine learning approaches inherently adapt to changing system dynamics, scale efficiently to high-dimensional problems, and can integrate diverse sources of information—from sensor streams to historical operational data—into unified decision frameworks. These advantages make ML particularly well-suited for addressing the evolving demands of power electronic systems where flexibility, predictive capability, and operational resilience are crucial. This dissertation advances machine learning applications in power electronics by proposing frameworks for optimal control, real-time in-situ health monitoring, predictive energy management, and hardware-accelerated on-edge inference in converter-dominated DC distribution systems. Validated through hardware-in-the-loop (HIL) and FPGA implementations, the methods deliver significant improvements in adaptability, computational efficiency, and fault resilience over traditional rule-based approaches. Chapter 1 addresses the lack of a unified view on ML's role in power electronics by performing a 2000-2024 bibliometric and thematic analysis, identifying the shift from heuristic and classical ML dominance to emerging methods like Graph Neural Networks (GNNs) and Generative Models (GenAI), and pinpointing open research gaps. Chapter 2 tackles the slow, manual converter design workflow by introducing a computer vision and OCR-based digitize-and-simulate pipeline that converts electrical schematics into a netlist for automated simulation, cutting design time from weeks to seconds. Chapter 3 responds to the challenge of limited topology innovation by building a structured graph-based dataset and applying Graph Variational Autoencoders and fine-tuned GPT-2 models to generate novel, high-performance converter designs. Chapter 4 addresses the difficulty of scalable, real-time grid health assessment by proposing an FPGA-accelerated, impedance-based monitoring method using neural network regression and classification to detect loading conditions and faults without distributed sensors or communication latency. Chapter 5 targets suboptimal energy scheduling in variable conditions through optimization-based techniques, i.e., mixed integer linear programming (MILP), and reinforcement learning-based techniques, i.e., Deep Q-Networks (DQN) agents that leverage intermittent renewable sources such as photovoltaic (PV), dynamic load profiles, and volatile electricity prices. Together, these contributions advance the state of the art from manual, expert domain knowledge-centric workflows toward fully autonomous DC distribution systems whose converters can be generated, verified, monitored, and dispatched by learning algorithm—paving the way for more resilient power converter-dominated DC distribution systems."]},{"key":"dc:format.mimetype","label":"Dc Format Mimetype","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["Machine Learning Enhanced Power Converter Design and Prognostic Health Monitoring in DC Distribution Systems"]}]}],"canonical_facts":{"dc:contributor.advisor":["Krishnamoorthy, Harish Sarma"],"dc:contributor.committeemember":["Nguyen, Hien V.","Fan, Lei","Shi, Jian","Rajashekara, Kaushik"],"dc:creator":["Bohara, Bharat 1991-"],"dc:date.accessioned":["2025-10-06T20:00:42Z"],"dc:date.issued":["2025-08"],"dc:description.abstract":["Power electronics has become an indispensable part of any modern electrical system, ranging from small to large-scale applications requiring efficient energy conversion at every power distribution stage. However, new challenges emerge as the converter-based systems grow in complexity, i.e., from individual power converter design and control to large-scale management of a series of interconnected converters in DC microgrids and DC distribution systems such as data centers, power grids, etc.. Though mature, well-established, and widely adopted, traditional model-based methods often struggle to address real-world challenges such as nonlinear system dynamics, uncertainties in the system parameters, and the growing complexity of interconnected converters in small to large-scale DC distribution systems. Moreover, real-time anomaly detection, health monitoring, and operational conditioning in such evolving systems require fast, adaptive, and scalable solutions to handle diverse operating conditions and unpredictable disturbances. Machine Learning (ML) provides a transformative toolset to tackle such inevitable real-world challenges prevalent in different power electronics domains. While traditional model-based approaches rely entirely on several stated assumptions and parameter estimations, data-driven models such as ML can capture complex system behaviors without solely relying on the idealized physics-based model, enabling predictive control, fault detection, and optimization under uncertainty scenarios. In converter design, ML can accelerate topology selection and parameter tuning; in operational control, it can facilitate adaptive decision-making and multi-objective optimization; and in large-scale DC systems, it can enable coordinated energy management across multiple nodes with varying priorities and constraints. ML can process high-frequency measurement data in real time for anomaly detection and health monitoring to identify subtle deviations indicative of incipient faults—capabilities that exceed the responsiveness of many traditional detection schemes. Compared to conventional techniques, machine learning approaches inherently adapt to changing system dynamics, scale efficiently to high-dimensional problems, and can integrate diverse sources of information—from sensor streams to historical operational data—into unified decision frameworks. These advantages make ML particularly well-suited for addressing the evolving demands of power electronic systems where flexibility, predictive capability, and operational resilience are crucial. This dissertation advances machine learning applications in power electronics by proposing frameworks for optimal control, real-time in-situ health monitoring, predictive energy management, and hardware-accelerated on-edge inference in converter-dominated DC distribution systems. Validated through hardware-in-the-loop (HIL) and FPGA implementations, the methods deliver significant improvements in adaptability, computational efficiency, and fault resilience over traditional rule-based approaches. Chapter 1 addresses the lack of a unified view on ML's role in power electronics by performing a 2000-2024 bibliometric and thematic analysis, identifying the shift from heuristic and classical ML dominance to emerging methods like Graph Neural Networks (GNNs) and Generative Models (GenAI), and pinpointing open research gaps. Chapter 2 tackles the slow, manual converter design workflow by introducing a computer vision and OCR-based digitize-and-simulate pipeline that converts electrical schematics into a netlist for automated simulation, cutting design time from weeks to seconds. Chapter 3 responds to the challenge of limited topology innovation by building a structured graph-based dataset and applying Graph Variational Autoencoders and fine-tuned GPT-2 models to generate novel, high-performance converter designs. Chapter 4 addresses the difficulty of scalable, real-time grid health assessment by proposing an FPGA-accelerated, impedance-based monitoring method using neural network regression and classification to detect loading conditions and faults without distributed sensors or communication latency. Chapter 5 targets suboptimal energy scheduling in variable conditions through optimization-based techniques, i.e., mixed integer linear programming (MILP), and reinforcement learning-based techniques, i.e., Deep Q-Networks (DQN) agents that leverage intermittent renewable sources such as photovoltaic (PV), dynamic load profiles, and volatile electricity prices. Together, these contributions advance the state of the art from manual, expert domain knowledge-centric workflows toward fully autonomous DC distribution systems whose converters can be generated, verified, monitored, and dispatched by learning algorithm—paving the way for more resilient power converter-dominated DC distribution systems."],"dc:format.mimetype":["application/pdf"],"dc:identifier.uri":["https://hdl.handle.net/10657/20708"],"dc:language.iso":["English"],"dc:subject":["Electrical engineering"],"dc:title":["Machine Learning Enhanced Power Converter Design and Prognostic Health Monitoring in DC Distribution Systems"],"dc:type":["Thesis"],"thesis:degree_discipline":["Electrical Engineering"],"thesis:degree_name":["Doctor of Philosophy"],"thesis:institution_name":["University of Houston"]},"updated_at":"2026-07-24T02:32:32Z"}