AI-Driven Design of Complex Inertial Fusion Energy Targets (Plasma Strand Project)

Inertial Fusion Energy (IFE) seeks to compress and heat fusion fuel using powerful drivers such as high-energy lasers. Each implosion must deliver high fusion gain to allow power generation. Direct-drive IFE, in which lasers directly illuminate the target, offers an attractive route to a high-repetition-rate fusion power plant. However, its design space is vast, highly non-linear and increasingly complex, as recent advances in target design (for example foams and graded dopants) and laser drivers (for example complex pulse shaping and increased bandwidth) open new opportunities that must be explored under reactor relevant conditions.

The central aim of the project is to identify high-gain implosion designs that maintain good performance in the presence of realistic system variability rather than optimising for idealised conditions alone. Finding resilient designs in this space demands a new generation of optimisation and surrogate modelling tools built on top of advanced radiation hydrodynamics simulations.

The student will use Chimera, the Centre for Inertial Fusion Studies’ (CIFS) state-of-the-art radiation–hydrodynamics code, to model IFE implosions including multi-dimensional effects. The project will advance the capabilities of Chimera to improve the accuracy of direct drive simulations. For example, the project student will implement reduced models of non-local heat flow, allowing a more accurate description of the laser-plasma interaction.

To enable optimisation, the student will develop new uncertainty-aware surrogate models which scale to large number of design parameters, for example, Gaussian process models with deep kernel learning. These surrogates will form the core of an optimisation framework, in which new simulations are chosen adaptively to refine the model in the most informative regions of design space, balancing one-dimensional surveys with targeted multi-dimensional simulations. Reactor-relevant constraints and sources of uncertainty, such as target manufacturing defects, target-injection and pointing offsets, and laser imprint, will be built into the surrogates and optimisation objective, so that the final designs are explicitly optimised for resilience to real-world operating conditions.

Through this work and training opportunities, the student will acquire skills at the interface of inertial fusion, numerical simulation and machine learning. Technical training will include: (i) radiation–hydrodynamics modelling using Chimera and related tools; (ii) modern scientific machine-learning methods; and (iii) high-performance computing and software development. In addition, the student will develop widely transferable skills in scientific writing (through key stage reports and papers), oral communication (through regular group meetings, seminars and conferences), problem solving and data analysis. They will be embedded within CIFS at Imperial College London, working closely with experts in inertial fusion and computational physics, and will have opportunities to engage with the wider fusion community (for example the United States national laboratories) and the AI-for-science community at Imperial.

Plasma strand students are based at University of York for the initial six months of the PhD, for the taught modules. During that first six months students will typically travel to undertake taught modules at all of the Fusion CDT partner universities. After the taught programme, the remainder of the PhD will be largely based at Imperial within the Centre for Inertial Fusion Studies, with opportunities to attend relevant international conferences.