Mechanical Engineering: Fully Funded Swansea University and UKAEA PhD Scholarship: AI and Inverse...

Funding providers: Swansea University's Faculty of Science and Engineering and UK Atomic Energy Authority (UKAEA)

Subject areas: Machine learning, computational mechanics, data, digital twin

Project start date: 

• 1 October 2022 (Enrolment open from mid-September)

Supervisors: 

• Professor Perumal Nithiarasu (Swansea)
• Dr Llion Evans (Swansea/UKAEA)
• Ms Michelle Tindall (UKAEA)

Aligned programme of study: PhD in Mechanical Engineering

Mode of study: Full-time

Project description:

The inside of a fusion reactor is one of the most challenging environments known about, with temperatures ranging from the hottest in the solar system (100,000,000 °C at the centre of the plasma) to the coolest (-269 °C in the cryopump) all within a few metres, coupled with electro-magnetic loads and irradiation damage. This has already been achieved for short periods of time at JET, the world’s largest fusion device located at Culham Centre for Fusion Energy (UKAEA), UK . But one of the greatest engineering challenges of the 21st century will be to construct a machine that can operate under these extremes routinely and produce commercially viable energy.

Video- UKAEA and Swansea University collaboration

UKAEA is building various prototypes for fusion components and assemblies. However, due to a range of loading conditions, the complexities involved in designing these make it challenging to create a design space. When following a conventional method, designers are often pinned to an extremely limited design space that may offer only one potential solution. By combining physics informed neural networks (PINNs) with inverse analysis, this project will investigate whether multiple solutions can be induced within a design space. An automated method of providing these multiple solutions to the designer gives materials, topology, and other parameter choices. To implement such a design space, we need to experiment with several different methods. Such outcome-based design will use state-of the-art machine learning methods, but the cost functions will be changed to induce multiple solutions. Also, perturbations and uncertainty in PINNs could be used to introduce multiple solutions. The resulting algorithms will be helpful to build digital twins of fusion systems and provide flexibility to designs. A proposed test case would be on a CHIMERA ‘sample under test’ (SUT), using the available simulation and test data; knowledge of the current design could then be used to explore multiple improved solutions.

The successful candidate will have a good undergraduate degree in a relevant subject, e.g., engineering, physics, or computer science. A postgraduate degree with relevant experience in the topics of this PhD is an added advantage. Previous specialisation in machine learning and/or computational mechanics will allow the student to rapidly start the work. The first year of the PhD will mostly be spent on testing novel machine learning methods for their suitability. The second year will allow the student to move into digital twin design and eventually leading to integration of the model into the workflow at UKAEA in the third year.

The project will be supported by access to high performance computing facilities  and funds to cover travel costs.