From Brittle to Ductile: Machine Learning 3D Fracture Simulations for Extreme Environments

This project aims to develop machine learning interatomic potentials (MLIPs) that accurately predict fracture and the brittle-ductile transition in tungsten for fusion reactors. Despite decades of experimental study [1], atomic-scale mechanisms remain elusive. You will build on recent 3D fracture simulations by our Max Planck partner [2] and state-of-the-art MLIPs for tungsten from a previous HetSys project [3], leveraging foundation models [4] to accelerate development.

Specific goals:

• Train MLIPs via active learning to capture brittle fracture and plastic deformation
• Perform large-scale atomistic simulations quantifying fracture toughness and critical temperatures
• Establish uncertainty quantification frameworks
• Validate predictions against experiments

[1] Gumbsch et al., Science 282, 1293 (1998)
[2] Hiremath et al., Comput. Mater. Sci. 207, 111283 (2022)
[3] Nutter et al., arXiv:2406.08368 (2024)
[4] Batatia et al., arXiv:2401.00096 (2024)

You will deliver:

• a publicly released, rigorously validated MLIP for tungsten, enabling the materials community to perform accurate fracture simulations;
• comprehensive datasets of fracture properties with uncertainty quantification suitable for engineering design;
• open-source simulation workflows and software tools for reproducible research;
• predictive understanding of crack-tip dislocation mechanisms governing the brittle-ductile transition;
• methodological advances in active learning for materials failure prediction applicable beyond tungsten.

Typically, HetSys PhD students publish around 1-2 academic publications in leading journals (although note this is not required). Your work will directly inform fusion reactor component design and establish new standards for ML-driven mechanical property prediction. You'll also build an international network through collaboration with Max Planck researchers.

Technical skills: Machine learning (neural networks, active learning, transfer learning), atomistic simulation methods (molecular dynamics, DFT), uncertainty quantification, high-performance computing, and scientific software development.

Domain expertise: Fracture mechanics, dislocation physics, materials thermodynamics, and fusion materials science.

Professional skills: Research software engineering (version control, testing, documentation), reproducible research practices, scientific communication, international collaboration, and project management.

Transferable skills: Data science, Python/C++ programming, statistical analysis, and problem-solving in complex systems.

These skills position you for careers in AI research, computational materials science, national laboratories (UKAEA, AWE-NST), tech industry, renewable energy, or academic research. HetSys training provides a foundation for these skills through dedicated courses and cohort activities.

We require at least a II(i) honours degree at BSc or an integrated masters degree (e.g. MPhys, MChem, MSci, MEng etc.) in a physical sciences, mathematics or engineering discipline. We do not accept applications from existing PhD holders.

If you are an overseas candidate please check here that you hold the equivalent grades before applying.

For postgraduate study in HetSys, the term “overseas” or “international” student refers to anyone who does not qualify for UK home fee status. This includes applicants from the European Union (EU), European Economic Area (EEA), and Switzerland, unless they hold settled or pre-settled status under the UK’s EU Settlement Scheme.