Masters Theses
Date of Award
12-2024
Degree Type
Thesis
Degree Name
Master of Science
Major
Nuclear Engineering
Major Professor
Jamie B. Coble
Committee Members
Xingang Zhao, G. Ivan Maldonado
Abstract
The pursuit of clean, cost-effective and reliable power has driven the development of advanced nuclear reactors, particularly for Small Modular Reactors (SMRs). This thesis presents a novel methodology for system-level health monitoring of SMRs using Neural Network (NN) Autoencoders and Fuzzy Logic Classification. The proposed approach addresses the challenges of fault detection and classification in dynamic operational environments, including both steady-state and load-following conditions.
Given the limited availability of operational data from SMRs, dynamic multi-physics simulations were utilized to generate synthetic data representative of an integral Pressurized Water Reactor (iPWR). The simulations were conducted using Modelica in the Dymola environment, focusing on fault scenarios affecting the Turbine Control Valve (TCV) and Feedwater Pump (FWP) systems. Two degradation modes were modeled onto the TCV system if we can differentiate them, these include a TCV leak and actuator failure. A general FWP fault was added to slightly decrease the flowrate throughout the simulation.
The developed health monitoring system employs a NN Autoencoder for fault detection by comparing real-time system data with reconstructed signals to identify anomalies. Statistical methods, including Simple Signal Thresholding (SST) and Sequential Probability Ratio Test (SPRT), were applied to the NN residuals to detect faults with high sensitivity. To enhance fault diagnosis, Fuzzy Logic Classification was integrated, providing a soft classification approach that reduces false positives and accurately distinguishes between multiple fault modes.
Results demonstrate the effectiveness of the NN Autoencoder and Fuzzy Logic Classification in identifying faults under both steady-state and load-following scenarios. The methodology proved capable of differentiating subtle faults and adapting to dynamic operational changes, highlighting its potential for improving the safety and reliability of SMRs. Additionally, the fuzzy logic system effectively reduced false alarms and provided valuable insights into the fault modes present, supporting more informed decision-making for plant operators.
This work contributes to the field of nuclear reactor health monitoring by demonstrating a scalable and adaptive fault detection and classification system. The integration of advanced neural network techniques with fuzzy logic enhances the ability to monitor complex reactor systems, offering a pathway towards more resilient and economically viable nuclear power operations.
Recommended Citation
Anderson, David J., "System-Level Health Monitoring of Small Modular Reactors: Neural Network Autoencoder Fault Detection and Fuzzy Logic Classification. " Master's Thesis, University of Tennessee, 2024.
https://trace.tennessee.edu/utk_gradthes/12837