PSAM 16 - Session Th02 Overview

PSAM 16 Conference Session Th02 Overview

Session Chair: Craig Primer (craig.primer@inl.gov)

Paper 1 DI318

Lead Author: Diego Mandelli Co-author(s): Congjian Wang, congjian.wang@inl.gov

From machine learning to machine reasoning: a model-based approach to analyze equipment reliability data
In current nuclear power plants (NPPs) a large amount of condition-based data which can be used to assess and monitor component health and performance. Assessing component health from such data can be performed with a large variety of methods. While the analysis of numeric data can be performed with several methods, the extraction of information from textual data remains a challenge. Currently employed natural language processing (NLP) methods do not really provide quantitative information that might be contained in IRs. In addition, the integration of numeric and textual data to identify possible causal relationships between data elements is still an unresolved challenge. This paper presents an approach to extract information from textual (e.g., incident or maintenance reports) and numeric data that relies on model based system engineer (MBSE) models. MBSE are diagrams designed to represent system and component dependencies (from both a form and functional point of view). In our approach, MBSE models emulate system engineer knowledge about component/system architecture. NLP methods are employed to perform syntactic and semantic analyses. Syntactic analysis analyzes the grammatical structure of a sentence while semantic analysis is designed to analyze the logic structure of a sentence. An innovative element of our approach is that semantic analysis uses MBSE models to identify links between textual elements. Similarly, numeric data is directly linked to elements of the MBSE models in order to map which functions are being monitored.

From machine learning to machine reasoning: a model-based approach to analyze equipment reliability data

In current nuclear power plants (NPPs) a large amount of condition-based data which can be used to assess and monitor component health and performance. Assessing component health from such data can be performed with a large variety of methods. While the analysis of numeric data can be performed with several methods, the extraction of information from textual data remains a challenge. Currently employed natural language processing (NLP) methods do not really provide quantitative information that might be contained in IRs. In addition, the integration of numeric and textual data to identify possible causal relationships between data elements is still an unresolved challenge. This paper presents an approach to extract information from textual (e.g., incident or maintenance reports) and numeric data that relies on model based system engineer (MBSE) models. MBSE are diagrams designed to represent system and component dependencies (from both a form and functional point of view). In our approach, MBSE models emulate system engineer knowledge about component/system architecture. NLP methods are employed to perform syntactic and semantic analyses. Syntactic analysis analyzes the grammatical structure of a sentence while semantic analysis is designed to analyze the logic structure of a sentence. An innovative element of our approach is that semantic analysis uses MBSE models to identify links between textual elements. Similarly, numeric data is directly linked to elements of the MBSE models in order to map which functions are being monitored.

Paper DI318 | |

Name: Diego Mandelli (diego.mandelli@inl.gov)

Bio:

Country: USA
Company: Idaho National Laboratory
Job Title: R&D Scientist

Paper 2 SC17

Lead Author: Sergio Cofre-Martel Co-author(s): Enrique Lopez Droguett eald@ucla.edu Mohammad Modarres modarres@umd.edu

Physics-Informed Neural Networks for Remaining Useful Life Estimation for Mechanical Systems
Prognostics and health management (PHM) has become a key instrument in the reliability community. Great efforts have gone into estimating systems’ remaining useful life (RUL) by taking advantage of monitoring data and data-driven models (DDMs). The latter have gained significant attention since they are model-independent and do not require previous knowledge of the system under study, known as black-box behavior. Nevertheless, DDMs developed for PHM frameworks are commonly tested on simulated or experimental datasets, which do not present the characteristics and intricacies of data collected from monitoring sensor networks in real systems. Furthermore, the black-box behavior hinders DDMs’ interpretability, and thus they are difficult to trust in the maintenance decision-making process. In this regard, physics-informed models have been implemented through hybrid models, which present significant improvements in accuracy and interpretability. Particularly, physics-informed neural networks (PINNS) have been proposed in deep learning (DL) to either solve or discover partial differential equations that govern a system. This paper presents an implementation of a PINN-RUL model to a case study from a real system. The system consists of a vapor recovery unit (VRU) at an off-shore oil production platform. Challenges when creating RUL labels based on maintenance logs are discussed. Results show that the PINN-RUL architecture is competitive with other traditional approaches, and it allows the interpretation of the system’s degradation dynamics through a latent variable.

Physics-Informed Neural Networks for Remaining Useful Life Estimation for Mechanical Systems

Prognostics and health management (PHM) has become a key instrument in the reliability community. Great efforts have gone into estimating systems’ remaining useful life (RUL) by taking advantage of monitoring data and data-driven models (DDMs). The latter have gained significant attention since they are model-independent and do not require previous knowledge of the system under study, known as black-box behavior. Nevertheless, DDMs developed for PHM frameworks are commonly tested on simulated or experimental datasets, which do not present the characteristics and intricacies of data collected from monitoring sensor networks in real systems. Furthermore, the black-box behavior hinders DDMs’ interpretability, and thus they are difficult to trust in the maintenance decision-making process. In this regard, physics-informed models have been implemented through hybrid models, which present significant improvements in accuracy and interpretability. Particularly, physics-informed neural networks (PINNS) have been proposed in deep learning (DL) to either solve or discover partial differential equations that govern a system. This paper presents an implementation of a PINN-RUL model to a case study from a real system. The system consists of a vapor recovery unit (VRU) at an off-shore oil production platform. Challenges when creating RUL labels based on maintenance logs are discussed. Results show that the PINN-RUL architecture is competitive with other traditional approaches, and it allows the interpretation of the system’s degradation dynamics through a latent variable.

Paper SC17 | Download the paper file. | Download the presentation PowerPoint file.

Name: Sergio Cofre-Martel (scofre@umd.edu)

Bio:

Country: USA
Company:
Job Title:

A PSAM Profile is not yet available for the presenter.

Paper 3 SA165

Lead Author: Sai Zhang Co-author(s): Fei Xu, Fei.Xu@inl.gov Zhegang Ma, Zhegang.Ma@inl.gov

Natural Language Processing-Enhanced Common Cause Failure Data Analysis
Nuclear power plants (NPPs) have a variety of operating records either routinely maintained or conditioned on incident occurrences. While some records are structured, the others are not. Analyzing records with unstructured data (e.g., narratives) can be challenging, not evaluating them would be a missed opportunity since they likely contain valuable operating experience. Given modern artificial intelligence and machine learning capabilities, it could be feasible and efficient to analyze unstructured NPP operating records. This paper presents an exploratory study of using natural language processing (NLP) to enhance common cause failure (CCF) data analysis for NPPs. The NLP-enhanced CCF data analysis attempts to (a) improve understanding of deep-level CCF failure propagation process and (b) compliment limited data pool of CCF events by analyzing non-CCF failure events and estimating their likelihood of evolving into CCF events. Both of these efforts will be implemented using the same approach developed in this study. The NLP-enhanced approach of analyzing a single CCF report includes the following steps: (1) breaking down the report by sentences and identifying causal relationships in each sentence, (2) analyzing and understanding logical relationships among different sentences, and (3) connecting the identified causal relationships in all sentences and forming a failure propagation network. A commonly-used, publicly-available source of unstructured records, licensee event reports, is used in this study. The approach will be applied further to analyze multiple equipment failure reports to evaluate whether a CCF event occurred, how likely an independent failure event can evolve to a CCF event, and to aggregate their failure propagation networks into one big network to draw statistical insights.

Natural Language Processing-Enhanced Common Cause Failure Data Analysis

Nuclear power plants (NPPs) have a variety of operating records either routinely maintained or conditioned on incident occurrences. While some records are structured, the others are not. Analyzing records with unstructured data (e.g., narratives) can be challenging, not evaluating them would be a missed opportunity since they likely contain valuable operating experience. Given modern artificial intelligence and machine learning capabilities, it could be feasible and efficient to analyze unstructured NPP operating records. This paper presents an exploratory study of using natural language processing (NLP) to enhance common cause failure (CCF) data analysis for NPPs. The NLP-enhanced CCF data analysis attempts to (a) improve understanding of deep-level CCF failure propagation process and (b) compliment limited data pool of CCF events by analyzing non-CCF failure events and estimating their likelihood of evolving into CCF events. Both of these efforts will be implemented using the same approach developed in this study. The NLP-enhanced approach of analyzing a single CCF report includes the following steps: (1) breaking down the report by sentences and identifying causal relationships in each sentence, (2) analyzing and understanding logical relationships among different sentences, and (3) connecting the identified causal relationships in all sentences and forming a failure propagation network. A commonly-used, publicly-available source of unstructured records, licensee event reports, is used in this study. The approach will be applied further to analyze multiple equipment failure reports to evaluate whether a CCF event occurred, how likely an independent failure event can evolve to a CCF event, and to aggregate their failure propagation networks into one big network to draw statistical insights.

Paper SA165 | | Download the presentation PowerPoint file.

Name: Sai Zhang (Sai.Zhang@inl.gov)

Bio: Dr. Sai Zhang works as a probabilistic risk and reliability analyst at Idaho National Laboratory, USA. Her research interests include probabilistic risk assessment, risk-informed analyses and applications, risk-cost optimization, and multi-criterion benefit evaluation for nuclear power plants. She has been a key investigator on a variety of U.S. Nuclear Regulatory Commission- and Department of Energy-sponsored projects including computational support for risk applications, Standardized Plant Analysis Risk (SPAR) modeling, risk-informed analyses for enhanced resilient plant systems, flood barrier testing strategy development, and exploration of artificial intelligence/machine learning in nuclear operating experience. She holds a Ph.D. degree in nuclear engineering from Tsinghua University, China.

Country: USA
Company: Idaho National Laboratory
Job Title: Probabilistic Risk and Reliability Analyst

Paper 4 AH324

Lead Author: Ahmad Al Rashdan Co-author(s): Roman Shaffer, romanshaffer@yahoo.com Edward (Ted) L. Quinn, tedquinn@cox.net

Fitness of computer vision machine learning in the regulatory framework for safety-related or risk-significant applications
With the advancements made in the field of artificial intelligence (AI) to date, significant potential exists to utilize AI capabilities for nuclear power plant (NPP) applications. AI can replicate human decision making, and AI decision making is usually faster and more accurate. For implementations that impact critical NPP applications (e.g., safety-related or non-safety systems that potentially affect overall plant risk), a deeper regulatory analysis of the AI methods is required. AI applied to NPP operations falls within the category of digital I&C (DI&C) because such applications involve digital computer hardware and custom-designed software that input plant data, execute complex software algorithms, and output the results to a system or licensed human operator to potentially provoke an action. For AI methods to be compliant with current regulatory requirements for DI&C, AI compatibility must be evaluated to identify AI gaps that may exist to prevent effective deployment of AI in NPPs. This work aims to evaluate how example AI technologies, specifically computer vision machine learning, align with the regulatory framework, and discusses some considerations if it is analyzed, modeled, tested, and validated in a manner similar to typical DI&C technologies.

Fitness of computer vision machine learning in the regulatory framework for safety-related or risk-significant applications

With the advancements made in the field of artificial intelligence (AI) to date, significant potential exists to utilize AI capabilities for nuclear power plant (NPP) applications. AI can replicate human decision making, and AI decision making is usually faster and more accurate. For implementations that impact critical NPP applications (e.g., safety-related or non-safety systems that potentially affect overall plant risk), a deeper regulatory analysis of the AI methods is required. AI applied to NPP operations falls within the category of digital I&C (DI&C) because such applications involve digital computer hardware and custom-designed software that input plant data, execute complex software algorithms, and output the results to a system or licensed human operator to potentially provoke an action. For AI methods to be compliant with current regulatory requirements for DI&C, AI compatibility must be evaluated to identify AI gaps that may exist to prevent effective deployment of AI in NPPs. This work aims to evaluate how example AI technologies, specifically computer vision machine learning, align with the regulatory framework, and discusses some considerations if it is analyzed, modeled, tested, and validated in a manner similar to typical DI&C technologies.

Paper AH324 | |

Name: Ahmad Al Rashdan (ahmad.alrashdan@inl.gov)

Bio: Ahmad Al Rashdan, Ph.D., is currently a senior research and development scientist at INL. Dr. Al Rashdan holds a Ph.D. in nuclear engineering from Texas A&M University, a M.Sc. in information technology and automation systems from Esslingen University of Applied Science in Germany, and a B.Sc. in mechanical engineering from Jordan University of Science and Technology in Jordan. He has more than 15 years of industrial and research experience. His experience includes automated work processes using AI methods and advanced analytics, online condition monitoring of nuclear systems, control-systems design and development, anomalies detection, and automated modeling and simulation. Dr. Al Rashdan is an active contributor to and organizer of several DOE events and scientific conferences. He authored or co-authored more than 60 technical reports and journal papers and 7 patent applications and is a senior member of the IEEE and a member of the American Nuclear Society.

Country: USA
Company: Idaho National Laboratory
Job Title: Senior Research and Development Scientist