Knowledge Graphs for Improved Decision Support for Nuclear Power Plant Asset Management

This PhD seeks to make a fundamental contribution to the emerging research discipline of Neuro-symbolic Artificial Intelligence, a sub-branch of the wider discipline of Artificial Intelligence. This advancement will be contextualized in engineering applications in nuclear power plant asset management, and specifically those concerned with the capture, codification of human (symbolic) reasoning and its subsequent utilization alongside large volumes of measurement data.

Nuclear power plants are part of a country's critical infrastructure supplying significant volumes of low carbon energy, that must be operated safely and efficiently. However, the amount of data generated by nuclear plants is vast and complex, making it difficult for operators and managers to make informed decisions. The application of AI & machine learning can help to address this challenge by providing operators and managers with models and algorithms which can support individual decision support tasks such as 1) Anomaly detection by defining models of normal operating behaviour and implementing algorithms which will flag to operators when deviations from normality occurs, 2) Fault diagnosis and classification by characterising behaviour by features drawn from sensor data and using this to determine the current condition of an asset, 3) Predicting remaining useful life of a piece of equipment by using sensor data to build a predictive model of future asset health.

However, in the vast majority of cases these data-driven approaches suffer from lack of transparency and explicability of their results, something which is essential to deployment within the nuclear sector. This PhD will focus on improving explicability in automated decision support.