Advanced data analysis and machine learning techniques for interpreting AGR reactor data

This is a multi-disciplinary project in the fields of machine learning and mechanical modelling with a funded PhD studentship. The student will be based in the School of Computer Science of the University of Manchester, and will closely interact with the EDF Energy R&D UK Centre to understand the industrial context.

Project Objective: The scope of this PhD project is to develop state-of-the-art data analysis and machine learning techniques in order to best interpret the data coming from the models and experiments of the AGR reactors. As they are producing an enormous amount of data, it is necessary to develop appropriate data analysis techniques to extract the relevant information, identify particular crack patterns of interest, and efficiently compare the different runs containing different input data.

Research Question: In the project, the main question to be researched is: How can machine learning models be used to effectively find correlation between inputs of interest (material properties, configuration of cracks, etc...) and outputs of interest (local deformations of the cores, peak forces, etc...)?

Approach: Multimodal and data fusion techniques in machine learning can be explored to jointly analyse data from models collected by different softwares and experiments. Moreover, the number of configurations of damaged core (position, shape and orientation of the cracks) being infinite, it is necessary to develop techniques that would enable to identify patterns of cracks that are of particular interest to support plant life extension, e.g. through analysing data statistics, regression and feature selection, and formulating relevant optimization problems, etc... The extrapolation of the results obtained would enable to find extreme scenarios, compare them to damage tolerances and eventually reduce the uncertainty and the degree of the conservatism of current procedures used in safety cases.

Novel Aspects: This project is closely related to the energy and engineering themes of EPSRC. The industrial partner EDF Energy team has developed the graphite core models under seismic loading and conducted experiments on a quarter scale simplified mock-up, to increase the understanding of the future behaviour of the AGR graphite cores beyond keyway root cracking. This provides a rich amount of heterogeneous data collected from multiple resources. Analysing and extracting hidden patterns from such data addresses new challenges in machine learning and data analytics. The project is of high impact, as the integrity of the graphite cores is critical, which are experiencing significant modifications with ageing due to oxidation, leading to weaker components, and irradiation, leading to greater slackness in the active. Late in the core operating life, graphite components are likely to crack, again leading to greater slackness in the active core and changed external load paths. Success of this project will facilitate the EDF Energy Generation team who is currently working on the extension of the life expectancy of their Advanced Gas-cooled Reactor (AGR) nuclear power plants, contribute to the underwriting of the safety cases for operation of the cores to their ultimate lifetimes, and enable EDF Energy to take informed lifetime investment decisions.