Projects / Programmes
Machine learning-based optimization of fusion reactor neutronics performance
| Code |
Science |
Field |
Subfield |
| 2.03.00 |
Engineering sciences and technologies |
Energy engineering |
|
| Code |
Science |
Field |
| 2.03 |
Engineering and Technology |
Mechanical engineering |
fusion reactor, neutron transport, tritium production, nuclear heating, machine learning
Researchers (1)
| no. |
Code |
Name and surname |
Research area |
Role |
Period |
No. of publicationsNo. of publications |
| 1. |
36329 |
PhD Aljaž Čufar |
Energy engineering |
Head |
2021 - 2024 |
651 |
Organisations (1)
| no. |
Code |
Research organisation |
City |
Registration number |
No. of publicationsNo. of publications |
| 1. |
0106 |
Jožef Stefan Institute |
Ljubljana |
5051606000 |
91,101 |
Abstract
Nuclear fusion has the potential to revolutionize the worldwide energy production due to its high availability of fuel, low physical and ecological footprint, and the wide availability of resources needed for operation. Fusion power plants will have to be designed with nuclear considerations in mind due to a high number of neutrons emitted during their operation. However, neutronics analyses represent one of the bottlenecks in the design and development of the fusion power plant. It can take weeks or even months to produce new models suitable for neutronics simulations. This often results in a rather narrow scope of neutronics analyses, leaving out many possible innovative alternative solutions that are significantly different to past designs. Parametric design and model preparation aim to reduce the human workload and thus open the possibility for studies with a wider and more diverse set of candidate designs. Furthermore, parametric modelling, automatic extraction of results, and fast evaluation of results are prerequisites for automating the design optimization. We see this as a next step in reactor design process that will be enabled by the continuous fast growth in available computational capabilities. While not presently feasible for complex and detailed models used in many of the state-of-the-art analyses, it is most likely already possible to apply these methods to simpler scoping studies. However, due to computational constraints resulting in a relatively low amount of possible test cases, it will likely not be enough to simply apply a standard optimization algorithm, e.g. genetic algorithm. Both a well-chosen definition of suitable criteria for the comparison of results and a way to take into account our understanding of the problem will be crucial. This demonstration of a new approach to fusion reactor design based on an automated optimization process will show a new way to reduce the human time needed for optimizing fusion power plant designs. We believe that combining novel ideas and modern approaches together with tried and tested methods is the only way to tackle the challenge of producing a viable design for a fusion power plant. The proposed research is a step in this direction.
