SPIDAR: System-level Physics-Informed Detection of Anomalies in Reactors

Publication Date

1-1-2023

Document Type

Conference Proceeding

Publication Title

Proceedings of 13th Nuclear Plant Instrumentation, Control and Human-Machine Interface Technologies, NPIC and HMIT 2023

DOI

10.13182/NPICHMIT23-41080

First Page

1124

Last Page

1133

Abstract

Nuclear power plants (NPPs) are home to many sensors that can be subject to various anomalies impacting the performance, safety, and reliability of NPPs. These sensor anomalies can arise due to degradation over time or as a result of various external factors. Anomaly detection models can be utilized to detect the presence of sensor errors in NPPs. Specifically, given the physical relationships present between sensed parameters in an NPP, physics-informed machine learning models can be developed to take advantage of the underlying physics of the system, described by physical equations, to ensure the model predictions remain physically consistent. As such, this study proposes SPIDAR: System-level Physics-Informed Detection of Anomalies in Reactors, a novel generative physics-informed system-level anomaly detection model for NPPs to detect sensor errors. Using data collected from a flow loop testbed, this study shows that SPIDAR can successfully detect anomalies present in an array of sensor data. Indeed, this study shows that SPIDAR outperforms state-of-the-art anomaly detection models, specifically a physics-uninformed GAN-based approach, highlighting the potential application of physics-informed machine learning for system-level anomaly detection in NPPs.

Funding Number

DE-NE0008978

Funding Sponsor

U.S. Department of Energy

Keywords

anomaly detection, generative adversarial networks, nuclear power plants, physics-informed machine learning

Department

Marketing and Business Analytics

Link to Full Text

Share

COinS