Considerations for Artificial Intelligence and Machine Learning in Nuclear Power: Interface Design and Experiment

Abstract

Automated technology integration is on the rise worldwide, and process control industries are no exception. Multiple factors are influencing this integration; for example, the commercial nuclear power industry is currently facing obsolescence issues and is evaluating opportunities to integrate automated technologies to remain economically competitive. One of the evolving disciplines of automation has been the introduction of artificial intelligence and machine learning models. These models are used for many purposes including processing large data sets, pattern discovery, and creating forecasts of various phenomena. Although there is significant opportunity for the inclusion of artificial intelligence and machine learning methods, nuclear utilities remain reluctant to adopt more advanced automation technologies due to some cultural barriers in their operations. The specific cultural barriers at hand are related to the overarching safety culture, which demands a level of skepticism of any system or indicator as well as a robust validation mechanism of any models relied upon. For this purpose, researchers at Idaho National Laboratory sought to understand how explainability and trust of predictive models might be integrated within a nuclear utility. This paper will describe preliminary results from the investigative effort for predictive modeling and how it can be integrated into nuclear utilities.