Lifetime extension programmes for existing civil and military nuclear assets necessitate improved monitoring, as well as modelling and prediction of equipment degradation throughout its lifetime. This project aims to develop a data-driven decision support system to enable faster, more effective management of nuclear assets.
Explaining the science
This project involves the development of data-driven decision support systems. By leveraging recent advances in machine learning and data science, these systems aim to integrate heterogeneous data streams from monitoring equipment, historical archives, and operator expertise in a semi-supervised fashion.
Supervised machine learning involves training a system to classify inputs into specific, known classes. The classic example is sorting email into spam. In the other extreme, unsupervised learning methods involving inferring conclusions from input data without labelled (i.e. cleaned up and classified) responses. Semi-supervised learning lies in between these two approaches, where one must draw inferences on a large amount of unlabelled data whilst only having access to a small amount of labelled data.
The data-driven decision support systems produced aim to provide nuclear asset operators and managers access to enhanced information to enable faster, more effective, and risk-informed decision making for safety critical operations.
Through recent developments in uncertainty quantification for machine learning algorithms, a stable and realisable support framework will be developed. This will encompass both diagnostic applications, i.e. identification of faults and characterisation of severity, and prognostic applications, where one is interested in predicting the evolution of future equipment health.
The work is part of a £4.4m EPRSC Prosperity Partnership funded project led by the University of Strathclyde and engineering firm Babcock.
The project also involves EDF Energy, Kinectrics Inc, Bruce Power, The Weir Group, BAM Nuttall, the University of Surrey, Cranfield University, and Imperial College.
The target application of this programme of research is specifically for decision support for operators of nuclear assets, providing risk assessed estimates of expected lifetime and degradation for key components within a nuclear energy environment.
The methods developed in this project would also be of use to asset operators in other industries, in particular for providing similar decision support for asset management within the oil and gas industry.
Furthermore, the foundational studies underlying the development of the methodologies being investigated will be of potential interest to the machine learning and computational statistics community.
The project is recruiting a postdoctoral research assistant to work on this project. Please contact the organiser for more information or to express interest.