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

Tokamak fusion reactors are a promising source of clean energy, but their structural materials operate under extreme conditions, including high temperatures and radiation. Radiation creates complex, evolving defect structures in the material, and predicting this damage is crucial for assessing component failure risk. High-fidelity atomistic simulations generate massive datasets describing millions of atomic positions. This project treats the challenge as one of dimensionality reduction and feature extraction, developing compressed “fingerprints” of these complex atomic systems. You will: Design and develop novel descriptors, with a focus on graph network-based methods (GNNs). Train and validate ML models on large, pre-existing datasets of damaged atomic structures. Model latent space dynamics from time-series data, moving from static snapshots to dynamic, predictive models. The outputs will feed into surrogate models for radiation damage, accelerating in-silico design and qualification of fusion reactor components. The work combines geometric deep learning, unsupervised representation learning, and time-series analysis. Industry Collaboration: This project includes an industry partner, UK Atomic Energy Authority (UKAEA), with an industry supervisor: Pratheek Shanthraj (pratheek.shanthraj@ukaea.uk)

Strong background in Mathematics, Computer Science, Data Science, or a related scientific domain
Experience or strong interest in machine learning, high-dimensional data, and predictive modelling

Apply via University of Manchester Application Portal under “PhD in Artificial Intelligence”
Required documents:
Final transcripts and certificates of all awarded qualifications
Interim transcripts of in-progress qualifications
CV
Supporting statement outlining motivation, relevant experience, and research skills
Contact details for two referees
English language certificate (if applicable)
Applications incomplete without all documents will not be processed