Explainable Artificial Intelligence for Identification of Human Errors in Nuclear Power Plants

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/NPICHMIT23-41185

First Page

1753

Last Page

1762

Abstract

Nuclear Power Plants (NPPs) may not be able to maintain their normal operating conditions owing to various causes such as human errors, mechanical defects, electrical defects, measurement defects, and external effects. Timely and accurate identification of incidents is important to restore NPPs to a stable state. However, the identification of abnormal operating conditions is difficult because of the existence of multiple scenarios. In addition, to implement mitigation actions rapidly after an incident occurs, operators must accurately identify an incident by monitoring the trends of many variables. The mental burden posed by this can increase human error and cause failure in identifying incidents. Failure in identifying incidents directly results in erroneous mitigation measures, which are detrimental to NPPs. Therefore, in this study, we develop artificial intelligence (AI) models to identify such errors, and thereby increase the chances of mitigating them, using the data collected from a physical testbed. We consider the faults occurring due to operators’ erroneous actions as “anomalies” to be detected. Thus, our goal is to detect as well as identify the various types of operator-level anomalies during interactions with the testbed, mimicking the context of an NPP. Additionally, having insights into what attributes contribute to the identification of anomalies provides us with intuition for informed decisions and crafting mitigation measures. As such, in this study, we leverage eXplainable Artificial Intelligence (XAI), including SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-Agnostic Explanations), to communicate potential operator errors back to the operators, allowing them to mitigate the errors. Our results show that the AI models developed and combined with XAI tools can explain the AI-prescribed decisions and potentially enable operators to better understand the sources and attributes associated with their errors.

Funding Number

DE-NE0008978

Funding Sponsor

U.S. Department of Energy

Keywords

anomaly classification, LIME, nuclear power plants, SHAP

Department

Marketing and Business Analytics

Link to Full Text

Share

COinS