Development of Signal Health Management Techniques for Anomaly Detection in Nuclear Reactors

다국어 초록 (Multilingual Abstract)

On-line Monitoring (OLM) recently became popular for nuclear industry applications. When analyzing industrial system’s data, bad data that is caused by electronics’ error such as; noise, outliers, and missing data can corrupt any diagnosing model and generate unreliable results, and that’s why bad data should be removed before implementing any decision step. Considering safety-related instrumentation in nuclear reactors, the redundant sensors are installed to alleviate the impact of bad data but their conditions should be more carefully monitored on-line, too, in order to maintain safety and operability. Therefore, in this study two tasks were performed; 1st task is to handle noise, outliers, and missing data at once, and 2nd task is to improve the on-line Cross Calibration (CC) averaging method applied to redundant sensors for the sake of diagnostics and system health management.
In the first task of this study, two new filters based on the well-known Moving Average and Moving Median techniques were introduced and proposed for cleaning noise, outliers, and missing data in a single technique, these filters are called “Cross Moving Average (CMA)” and “Cross Moving Median (CMM)”. Comparing these filters performance with the conventional Autoregressive model, Moving Average (MA), and Moving Median (MM) methods and evaluating them under some statistical evaluation criteria, the result of CMA and CMM showed a competitive performance.
The second task in this study was performed in two paths; one is improving Parity Space Averaging technique, which is an important technique among CC techniques, by involving a new weighting factor based on distance instead of the weighting factor based on sensor’s accuracy that was implemented in the Instrument Calibration and Monitoring Program (ICMP) by Electric Power Research Institute (EPRI). the other is to introduce another two new weighting factors; one based on trend consistency, and one based on the attenuation of outliers, to integrate the advantages of all conventional CC averaging techniques. When implemented on redundant signals from reactor facilities, these weighting factors effectively worked to isolate signals’ anomalies comparing to the method implemented in the ICMP.

목차 (Table of Contents)

• 1. General Introduction 1
• 1.1. Background 1
• 1.2. Motivation 3
• 1.3. Scope & Objectives 5
• 1.4. Dissertation Organization 6
• 2. Data Cleaning 8
• 2.1. Introduction 8
• 2.2. Literature Survey 11
• 2.2.1. Data Problems 11
• 2.2.2. Data Processing 16
• 2.2.3. Statistical Validation 20
• 2.3. Methodology 23
• 2.3.1. Cross Moving Average (CMA) filter 24
• 2.3.2. Cross Moving Median (CMM) filter 24
• 2.3.3. Evaluation criteria 25
• 2.4. Results and Discussion 32
• 2.4.1. Validation of missing data recovery 32
• 2.4.2. Descriptive Statistics analysis 34
• 2.4.3. Evaluation analysis 37
• 2.5. Conclusions 44
• 3. Cross Calibration Averaging Techniques 47
• 3.1. Introduction 47
• 3.2. Literature Survey 50
• 3.2.1. On-line Cross Calibration Averaging 50
• 3.2.2. Deviation Limits 56
• 3.3. Methodologies 60
• 3.3.1. Modified Parity Space Averaging approach (MPSA) 60
• 3.3.2. Integrated Cross Calibration Averaging (ICCA) 61
• 3.3.3. Implemented models 67
• 3.3.4. Applied Deviation limits 69
• 3.3.5. Drift Index (DI) 70
• 3.4. Results and Discussion 71
• 3.4.1. Validation using research reactor data sets 71
• 3.4.2. Validation using Nuclear Power Plant data sets 82
• 3.5. Conclusion 91
• 4. General Conclusions 93
• 4.1. Conclusions 93
• 4.2. Findings 96
• 4.3. Recommendation for future work 97