Unsupervised machine learning techniques for fault detection and diagnosis in nuclear power plants

Abstract

Nuclear power plants have proved their importance in the energy sector by generating clean and uninterrupted energy over decades. Moreover, nuclear power plants (NPPs) are large-scale and complex systems with potential radioactive release risks. Hence, it is important to achieve the highest standards of safety by designing a reliable process monitoring system. Due to the massive amount of monitoring data that are collected and stored in modern NPPs, it is difficult to extract necessary information about the actual plant state in a timely and accurate manner. Machine learning techniques play a vital role as tools for data mining that help in decision-making and support the operators in the process industry. This paper presents an integrated framework for fault detection and diagnosis in NPPs using unsupervised machine learning techniques where there is no prior knowledge is required. In this framework, the faults are first detected using the principal component analysis (PCA) approach by different fault detection indices. Second, the multivariate contribution plots (MCPs) methods are used for fault diagnosis. The data collected from Personal Computer Transient Analyser (PCTRAN) simulation is utilized to demonstrate the efficiencies of all the discussed methods.