Predictive analytics for maintenance activities in nuclear power plants: a feasibility study

Abstract

Nuclear power plants are known for their use of legacy systems and processes. As plants age, the amount of maintenance increases while resources remain finite, leading to unwanted delays, affecting the health of assets and increasing costs. To aid in the modernization and digitization of nuclear power plants, this work explores data driven methods, including statistical and machine learning techniques to predict target variables. Representative Naval Propulsion Plant data with variables similar to that in the nuclear industry are used as nuclear data is not available in the public domain. Experimental results confirm target variables can be predicted with relatively high accuracy, with Deep Learning methods harbouring the lowest relative error. Two frameworks are developed based on results to showcase how predictive analytics can be used in nuclear power plant maintenance. This work is a proof of concept informing stakeholders that data driven approaches are viable in reducing maintenance delays.