Data Fusion of Numerical and Textual Equipment Reliability Data: A Knowledge-Graph Based Approach

Mandelli, D.; Wang, C.; Cogliati, J.

Permalink

https://hdl.handle.net/2142/121801 Copy

Description

Title

Data Fusion of Numerical and Textual Equipment Reliability Data: A Knowledge-Graph Based Approach

Author(s)

• Mandelli, D.
• Wang, C.
• Cogliati, J.

Issue Date

2023

Keyword(s)

• Reliability
• Natural Language Processing (NLP)
• Data fusion
• Model Based System Engineering (MBSE)

Abstract

Nuclear power plants are collecting large amount of data equipment reliability (ER) elements that contain information about the status of component, assets, and systems. Such data might be in the form of on-line monitoring data (e.g., pump vibration data, or pump mass flowrate), surveillance and testing data (performed by plant operators at regular intervals), condition reports (which typically contain anomalous conditions), and maintenance reports (which indicate operations performed to restore component or asset health). All these data elements precisely record assets/systems performance/health throughout their lifecycle. In addition, such data have the potentials to provide insights to system engineers about the presence of anomalous behaviors or degradation trends, the possible causes of such behaviors and trends, and identify in advance their direct consequences. However, several challenges have proven to be roadblocks to reach such potentials; while some are technical in nature (i.e., data is often distributed over several physical servers/databases), some others are conceptual: data elements have different format (e.g., numeric, textual), and measured values have different scales (e.g., vibration spectra, oil temperature). This paper directly tackles these challenges and it focuses on the integration of nuclear and textual data elements with the goal of assisting plant system engineers to analyze ER data. This task is initially performed by pre-processing the data by: 1) extracting knowledge from textual data using natural language processing (NLP) methods, and 2) quantifying system/asset/component health from numeric data. Then we employ model based system engineering (MBSE) models of plant, systems and assets to identify their architecture and functional, i.e. cause-effect, relations. These models are translated into graph structures where each element of the graph represents either the “form” of a system, asset and component or their supporting “function”. Data elements are then associated with a single MBSE graph element based on their nature. This bonding of MBSE models and data elements constitutes the first of its kind knowledge graph of a nuclear power plant. At this point data elements are organized in a structured way such that system engineers can identify cause-effect patterns between data elements and act accordingly.

Type of Resource

text

Language

eng

Handle URL

https://hdl.handle.net/2142/121801

Owning Collections