Using Multivariate Statistical Surrogate Models of Displacement Cascades To Simulate High-dose Irradiation Damage

Details

The materials of a fusion reactor must survive for many years in an incredibly harsh environment. This includes constant bombardment by high energy neutrons, which smash into the materials, rearranging the atoms in our carefully engineered metallic alloys. We need to understand how this process of irradiation damage will change the properties of the alloys and perhaps limit their useful lives.

Direct experimental tests of performance under high-energy neutron bombardment are at best time consuming and expensive and at worst impossible. Experimental characterisation techniques struggle to see damage at the smallest scale (individual atoms) and cannot capture the rapid processes of initial damage (less than nanoseconds). This is one area where computer simulations of materials behaviour must be our eyes on these critically important processes.

But simulations have their limitations too. We can simulate individual damage events, involving millions of atoms, with good representations of the true physics. But these simulations are expensive and cannot directly treat both the small time and length scales of the individual damage events and the long term accumulation of damage over many months. To get around this limitation, recent work has focused on encapsulating the detail of the individual damage events, as derived from expensive simulations, in cheaper surrogate models.

In this project you will continue this work, first creating a database of irradiation damage simulations in iron, using classical molecular dynamic simulations. Next you will characterise the populations of defects produced by these events, developing statistical descriptors of the damage using both traditional approaches and deep-learning methods. These descriptors will then become the basis of surrogate models that you will use to explore high-dose damage in Fe. You will consider not only damage in perfect crystal bulk Fe, but also explore how extended defects, such as grain boundaries, phase boundaries and dislocations, affect the statistics of defect production, to produce a surrogate modelling platform capable of treating these key features of the microstructure of real steels for fusion power.

This project is focused on materials simulation and statistical modelling. It will suit a student interested in computational work and data analysis. Training will be provided in all the core techniques and in the materials background to the project. You’ll need to do some coding, but only in an easy-to-use scripting language like Python, which you can learn as part of the project. There won’t be any hard-core software development. The project would particularly suit you if you enjoyed the mathematical elements of your undergraduate course, whether that was in physics, materials, engineering or a related discipline.

The project will be mainly based in Sheffield. There will be opportunities for travel to CCFE in Culham and for international travel for conferences and collaboration. The extent of travel will be tailored to the needs and desires of the student to make sure the project is as accessible to all, regardless of personal circumstances.