Advancing Fusion Energy with AI—Developing Novel Compressed Representations for High-Dimensional Physics Data

This fully funded PhD project at The University of Manchester, in partnership with the UK Atomic Energy Authority (UKAEA), aims to advance fusion energy research by developing novel AI-driven compressed representations for high-dimensional physics data. Tokamak fusion reactors, a promising source of clean energy, require structural materials that can withstand extreme conditions, including high temperatures and intense radiation. Radiation bombardment leads to complex, evolving defect structures within these materials, and predicting such damage is crucial for assessing the risk of component failure. High-fidelity atomistic simulations can model these damage events, but the resulting data—describing the positions of millions of atoms—is too large and high-dimensional for direct use in predictive models. The core challenge is dimensionality reduction and feature extraction: creating efficient, compressed 'fingerprints' of atomic structures that enable actionable insights. The project treats the evolution of material damage as a large-scale pattern recognition problem, focusing on developing machine learning algorithms capable of fingerprinting vast atomic datasets and capturing complex, long-range atomic neighborhoods. You will work at the intersection of geometric deep learning, unsupervised representation learning, and time-series analysis. Key tasks include designing and developing novel descriptors, especially using graph neural network (GNN) methods, training and validating models on a large database of damaged atomic structures, and modeling latent space dynamics using time-series data. The compressed descriptors you create will be essential for building fast, efficient surrogate models for radiation damage, directly supporting the in-silico design and qualification of new fusion reactor components. The project offers access to extensive datasets and collaboration with both academic and industry supervisors. Applicants should have a strong background in mathematics, computer science, or a relevant scientific domain. The program is part of the AI UKRI CDT, offering full funding, home tuition fees, and a tax-free stipend at the UKRI rate (£20,780 for 2025/26), with a start date in September 2026. The University of Manchester is committed to equality, diversity, and inclusion, encouraging applications from all backgrounds and offering flexible study options. To apply, submit your application via the university portal, including all required documents and referee details. For further details, see the official FindAPhD listing.