Doctoral Dissertations

Orcid ID

https://orcid.org/0000-0001-8565-4094

Date of Award

12-2025

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Major

Energy Science and Engineering

Major Professor

Jamie B. Coble

Committee Members

Pradeep Ramuhalli, Bing Yao, Xingang Zhao

Abstract

Ensuring the reliability, safety, and cost-effectiveness of nuclear power plants requires advanced methods for fault detection, diagnostics, and prognostics that can operate under changing conditions and across multiple system scales. This dissertation develops and evaluates data-driven frameworks that span anomaly detection, component-level fault diagnosis, and system-level prognostics for critical nuclear systems. The first study introduces a unified framework that integrates statistical, model-based, and data-driven methods to detect and investigate anomalies in complex engineered systems, demonstrating robust and interpretable detection on SMART valve systems. Building on this, the second study evaluates one-class learning approaches for fault detection in the circulating water system of a nuclear power plant, showing that embedding-based models such as Deep Centered Embedding (DCE) generalize effectively under distributional shifts. The third study addresses fault detection and diagnostics in Fine Motion Control Rod Drives (FMCRDs), an underexplored subsystem critical for reactor control, and demonstrates the use of neural architectures for identifying and localizing electrical and mechanical faults. Shifting focus to prognostics, the fourth study investigates data-driven prediction of condenser tube fouling using simulation data, finding that recurrent neural networks, particularly Long Short-Term Memory (LSTM) models, provide accurate and noise-resistant estimates of remaining useful life (RUL). Finally, the fifth study extends prognostics to the system level by introducing a novel topology-aware health index and comparing machine learning approaches, with Graph Neural Networks (GNNs) emerging as the most effective for capturing inter-component dependencies and delivering reliable system-level RUL predictions. Collectively, these studies contribute a comprehensive methodology for advancing predictive maintenance in nuclear power plants, from anomaly detection to system-level prognostics, and provide a foundation for enhancing the resilience and sustainability of critical energy infrastructure.

Available for download on Tuesday, December 15, 2026

Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS